b5:T3142,PROMPT HEADER: Create a 15-second horizontal 16:9 photorealistic live-action absurdist video in one uninterrupted first-person take at 30 fps. The viewer is a student quietly filming on a phone from a seated position near the back of an American high-school classroom with a practical private-school dress code. Keep a fixed, slightly wide-angle composition with only small natural handheld tremors. Rows of desks fill the left and center, all facing the distant chalkboard at the front, and an enormous wheeled cannon occupies the right foreground beside tall windows. No cuts, zooms, reverse angles, or time jumps. Everyone treats the absurd sight as routine. CAMERA AND CLASSROOM LAYOUT: Build a recognizable American classroom with painted cinder-block walls, fluorescent ceiling lights, a wall clock over a large green chalkboard, a side bulletin board with English class notices, a U.S. map, slightly worn tile flooring, metal-framed student desks, open textbooks, notebooks, pencils, and piles of classwork. Tall white-framed windows run along the right wall, with pale floor-length curtains. Outside is an empty American campus athletic field and a low brick school building with distant trees. Keep the exterior sunlit and clearly visible through the windows. Camera placement and seating direction are nonnegotiable: the lens is BEHIND the ordinary students and looks toward the front wall. Every ordinary desk has its writing edge and chair oriented toward the chalkboard, with students facing away from the camera. The lens mostly sees backs of heads, shoulders, and uniform jackets. A small side profile may appear when someone turns. Never show a row of ties, chests, and full faces facing the lens while the teacher writes behind those students. Only the girl operating the cannon is seated sideways by the right window. The nearest student on the left bends over a workbook. STUDENTS AND UNIFORM: All people have distinct Western faces and stable identities. The older teenage students wear one coherent contemporary American private-school uniform: white or pale-blue polo or collared shirts, charcoal chinos, knee-length navy-and-gray plaid skirts, and optional navy sweaters or blazers with a modest school crest. A few students wear loosened navy ties; most do not. The clothing is practical and casually worn. The unseen operator filming the scene wears the same uniform. Keep all ordinary students at their assigned desks. They write, turn pages, or make small natural movements while facing the chalkboard. No ordinary student enters the aisle before the cannon fires. TEACHER CONTINUITY: There is exactly ONE teacher. She is an adult Western woman with a short dark-brown bob, a plain white short-sleeved button-down blouse, and dark trousers. At the beginning she stands at the distant chalkboard with her back or side to the camera and writes with chalk. She is the SAME woman who later leaves the board, walks down the aisle, speaks to the class, notices the phone, and confronts the student filming. Keep her appearance consistent and never add a second adult. She writes, pauses, scolds the class, then notices the phone. CANNON GEOMETRY AND OPERATOR: Lock the cannon geometry before any action begins. The huge dark-steel carriage, two thick black rubber tires with solid enclosed hubs, rear breech, and controls are INSIDE the classroom in the lower-right foreground. A long pale-gray metal barrel runs diagonally UP AND TO THE RIGHT from that breech, crosses a visibly open section of the right-side window, and continues OUTSIDE toward the empty field. Its muzzle lies beyond the outer wall near the upper-right edge of the frame. Show one continuous, rigid barrel from interior breech to exterior muzzle in every frame, including during recoil. This exact relationship is the most important spatial constraint: breech inside, barrel through the open right window, muzzle outside. Its line of fire points away from the classroom. It never angles left across the desks or toward the teacher, students, or camera. It never ends inside the room or passes through closed glass. It is a modern field gun with broad tires and enclosed hubs, never an antique spoked-wheel cannon. Never shorten or reverse the barrel. A Western teenage girl with long dark hair sits on a low seat at the rear-left side of the cannon, between it and the nearest desks. She is shown in profile, with her hands near the rear controls, wearing the same school's white shirt, navy uniform layer, and plaid skirt. She stays seated in this position throughout. Before firing she makes small precise adjustments; afterward she remains composed. She never leaves her seat. CONTINUOUS SHOT PLAN: 0.0-4.5 seconds — Quiet study hall: Hold the established wide first-person view. At the front, the only teacher is still writing on the chalkboard, her chalk hand moving in short strokes. She may reach higher but does not turn, speak, or leave. Students, viewed from behind, lean over their books, write, and turn pages while facing the chalkboard. The girl beside the cannon adjusts its rear mechanism and settles her hands. The full barrel remains unmistakably aimed out of the right window. No one enters the aisle or approaches the lens before the shot. 4.6-4.9 seconds — One outward discharge: The seated girl fires the cannon exactly ONCE, outward toward the unoccupied field. Her small control movement leads to one abrupt report. Show a split-second warm flash at or just beyond the EXTERIOR muzzle, a brief recoil and vibration through the fixed carriage, then a restrained puff of pale smoke at the open window. The nearby long curtain snaps upward and inward, twists in the moving air, and begins to settle. Normal daylight returns immediately. Both solid-hub rubber wheels stay planted and the long barrel retains its rightward outdoor direction. Show no indoor fire, shown impact, or second discharge. 5.0-8.8 seconds — Deadpan aftermath: Show the aftermath of that single shot. The curtain continues to sway and gradually falls back. Thin haze drifts along the right windows and starts to disperse. Its continuing motion must not read as another cannon shot. A few students briefly lift their heads or shift in their seats, while most keep working and quickly resume writing. The girl at the controls stays calmly seated. At the far front, the teacher stops writing, lowers her chalk hand, and turns from the board after hearing the report. She remains near the front during this immediate aftermath and is annoyed by the noise. The room stays fixed. 8.8-10.5 seconds — Teacher's first reprimand: The SAME teacher faces the room and says in natural American English, firmly and matter-of-factly: “This is study hall. Keep it quiet.” She glances over the desks, then steps away from the chalkboard into the center aisle. Start her walk visibly at the front. She must gradually grow larger as she passes successive desk rows. Show a continuous normal-paced walk from the front. The operator remains seated and the barrel remains outside. 10.5-13.9 seconds — She notices the phone: The teacher walks farther down the same aisle, with desks on her left and the cannon operator near her right as she approaches the phone viewpoint. Her gaze first scans the room and then fixes directly on the lens when she notices the unseen student recording. Her brows tighten slightly. Her face changes from general irritation about the noise to direct disapproval of the phone. Address only the person behind the camera, not the girl at the cannon. Say once, clearly and with natural lip synchronization: “And you—why is your phone out? Put it away.” The unseen student does not reply; the phone itself need not appear. Keep her walk continuous from the chalkboard into the foreground. 13.9-15.0 seconds — Hand covers the lens: Now close enough for her face and upper body to be clearly recognized, the teacher leans toward the lens, raises one hand from her side in a visible motion, and extends her open palm toward the student's phone. Her face remains briefly visible behind the growing hand. The hand fills the frame naturally as it reaches the lens, and the image darkens because the palm physically covers it. Finish there. This is the only full lens obstruction in the video: no early hand, no disembodied hand, and no hidden cut behind it. The camera may shift a fraction in the student's grip at the final touch but never changes viewpoint. PERFORMANCE, AUDIO, AND TONE: Keep the pacing fluid rather than freezing each time interval into a separate shot. The teacher's chalk writing leads into her pause, turn, spoken warning, uninterrupted walk, phone reprimand, and final reach. The girl's adjustment leads into one outward cannon discharge, then one continuous curtain-and-haze aftermath. Use modest classroom ambience, chalk and pencil sounds, one short heavy report, moving fabric, footsteps, and only the teacher's two short English lines. Keep both lines intelligible. No other dialogue, music, captions, titles, or logos. The final tone is deadpan: the teacher objects to noise and a phone while the enormous cannon remains beside students who are quietly doing their work. Preserve all character identities, uniforms, props, and spatial relationships from the first frame to the last. VISUAL AND CONTINUITY RULES: Photorealistic live-action classroom comedy with natural daylight from the right, practical fluorescent light, restrained handheld motion, realistic smoke and fabric movement, stable faces, stable uniforms, and physically coherent props. Keep the classroom architecture fixed. Keep the cannon barrel rigid and continuous from breech to exterior muzzle. Keep the window section open where the barrel passes through. Keep the cannon operator seated and the teacher's walk physically continuous. Do not turn the scene into an action film, war scene, horror scene, or video game. The absurdity comes from the students and teacher treating the cannon as ordinary while reacting only to noise and phone use. NEGATIVE PROMPT: Second teacher, extra adults, duplicate teacher, teacher identity drift, teacher teleportation, teacher starting near the camera, teacher appearing without walking, teacher addressing the cannon operator, teacher facing the wrong direction, students facing the camera at ordinary desks, rows of full faces toward the lens, desks facing away from the chalkboard, reversed classroom layout, cannon aimed left, cannon aimed at people, cannon aimed at camera, barrel ending inside classroom, shortened barrel, reversed barrel, broken barrel, barrel through closed glass, missing open window, floating cannon, antique spoked wheels, missing tires, detached carriage, changing cannon geometry, cannon recoil changing its direction, second firing, repeated muzzle flash, indoor muzzle flash, shown impact, fire inside classroom, explosion, dangerous debris, cannon operator standing, operator changing seats, operator leaving frame, students abandoning desks, extra cannon, duplicate students, cloned faces, unstable identities, changed uniforms, missing chalkboard, missing windows, missing curtains, dark exterior, unreadable classroom signs, readable brand names, captions, subtitles, titles, logos, watermark, music, extra dialogue, missing dialogue, bad lip synchronization, early hand obstruction, disembodied hand, hidden cut behind hand, zoom, reverse angle, time jump, camera viewpoint change, excessive camera shake, excessive smoke, excessive motion blur, horror lighting, war-movie tone, videogame render, low resolution, blurry faces, distorted hands, extra fingers, extra limbs, broken anatomy, unnatural curtain motion, sound mismatch. Final Summary Create one complete 15-second horizontal 16:9 photorealistic live-action absurdist classroom video at 30 fps. Use Reference Video only for fixed first-person framing, subtle handheld movement, natural pacing, and deadpan comedy timing. The written prompt controls the classroom, one consistent teacher, uniformed students viewed from behind, the seated cannon operator, the cannon's interior-to-exterior geometry, one outward shot, the teacher's two lines, her continuous walk, and the final palm covering the lens. The finished video should make the joke legible without changing the room's ordinary routine: a cannon fires outside the classroom, but the teacher's priorities are keeping study hall quiet and stopping the student from filming. Preserve the spatial logic, physical continuity, two short English lines, one heavy report, restrained aftermath, and final lens obstruction. b6:T1e79,GLOBAL PROMPT HEADER: Entirely original premium photorealistic live-action fantasy adventure. Cinematic horizontal 16:9, 1080p, 24 fps, restrained 35mm anamorphic camera, physically grounded action, natural volumetric light, realistic metal, stone, cloth, rain, vegetation, and skin. Character1 and Character3 are the same hero at different progression stages. Character4 remains the same fencer; Character2 is the only commander; Character6 is the only lancer. Preserve faces, hair, costumes, weapon counts, and screen direction across cuts. Use only original medieval-European floating cities, gothic towers, plazas, bridges, ruins, cathedrals, and forest sanctuaries. No Japanese roofs, franchise names, copied emblems, readable words, letters, numbers, subtitles, captions, logos, watermarks, modern HUD, guns, blood, gore, extra limbs, duplicated people, face drift, costume drift, morphing, or teleportation. Any magical panel is translucent, abstract, icon-only, and unreadable. Use original score and synchronized environmental sound; spoken lines are Japanese with no subtitles. CLIP 01 - 00:00-00:18 - ENTERING THE WORLD: Attach @Character1 and @Reference Video. In a modern bedroom at night, use Character1's face in a charcoal T-shirt as he puts on a plain black headset. Cut through blue-white particles to a sunlit floating medieval plaza, where he materializes in the exact Character1 beginner gear with one sword and looks upward in wonder. CLIP 02 - 00:18-00:34 - STARTING CITY LOCKDOWN: Attach @Character1, @Character5, and @Reference Video. Character1 stands among a controlled crowd in the same plaza. Ice-blue icon-only panels become muted amber-red warning shapes; no text. Use a close reaction, a high aerial, and a wide crowd exit as he grips his sword strap and chooses a direction. CLIP 03 - 00:34-00:50 - FIRST ALLIES: Attach @Character1, @Character4, @Character6, @Character5, and @Reference Video. At a rain-darkened gate, Character1 meets Character4 and Character6. A small wooden training construct attacks: Character6 braces her spear, Character4 parries once, Character1 destroys it with one measured sword strike. They agree in short Japanese lines to travel together and cross the bridge. CLIP 04 - 00:50-01:08 - PROGRESSION: Attach @Character1, @Character4, @Character6, @Character5, and @Reference Video. Use five clean cuts: wet bridge, stone stair, ruined-courtyard training, campfire rest, and dawn outpost. Character1 has one sword, Character4 one rapier, and Character6 one spear throughout. End on the group approaching a blue-lit ruin. CLIP 05 - 01:08-01:26 - BLUE RUIN RAID: Attach @Character1, @Character4, @Character6, @Character5, and @Reference Video. In a cold-blue ruined cathedral, one horned obsidian mineral guardian drags a stone cleaver toward the expedition. Character1 protects the women behind a pillar, opens one icon-only panel, selects a second sword, then draws two distinct swords. Character4 warns him; he answers that he will hold the line. CLIP 06 - 01:26-01:46 - DUAL-BLADE VICTORY: Attach @Character1, @Character4, @Character6, and @Reference Video. Six readable combat cuts: evade, crossed-blade block, Character6 spear brace, Character4 ankle strike, frontal sprint, then a final X-shaped impact into the guardian's chest fracture. It shatters into blue-white mineral particles; Character4 catches the exhausted hero. Keep both swords distinct. CLIP 07 - 01:46-02:04 - FOREST SANCTUARY: Attach @Character3, @Character4, and @Reference Video. In a quiet lake forest with a small timber-and-stone cottage, use the advanced Character3 coat but no swords. Character4 wears an original cream blouse and muted rose skirt. Aerial, hand-holding walk, two bowls on a table, then a sunset dock embrace. No other people. CLIP 08 - 02:04-02:22 - RETURN TO THE FRONT: Attach @Character3, @Character4, @Character6, @Character5, and @Reference Video. Hard cut to a stormy high-level bridge. Character3 has exactly two sheathed swords, Character4 is armored again, and Character6 carries one spear. Use a low approach, floating-ruin aerial, silent command gesture, icon-only two-sword panel, and crimson tower doors opening. CLIP 09 - 02:22-02:40 - COMMANDER REVEAL: Attach @Character2, @Character3, @Character4, @Character6, and @Reference Video. In a cold upper hall, Character3 makes one controlled sword thrust toward Character2. A thin amber geometric barrier stops it before armor. He asks in Japanese whether the commander made this world; the commander confirms. Blue light changes to crimson and the hall becomes a red gothic cathedral. End with the rectangular shield on the floor. CLIP 10 - 02:40-02:58 - FIRST DUEL: Attach @Character2, @Character3, and @Reference Video. No other people may appear. Make at least ten fast readable shots of dual-sword attacks, shield parries, footwork, sparks, a low slide, and a weapon lock. Character2's shield stays large and rectangular. End with one shield impact knocking Character3 down; Character2 raises his sword for a finishing strike. No blood. CLIP 11 - 02:58-03:18 - INTERCEPTION: Attach @Character2, @Character3, @Character4, and @Reference Video. Start behind fallen Character3's shoulder. Character2 swings downward; only now does Character4 enter, block the attack with rapier and body, then look back and tell him to live. No blood or wound. Her rapier drops and she dissolves fully into pale-gray ash in the crimson wind. Character3 remains staring at the empty space. CLIP 12 - 03:18-03:38 - BREAK THE SHIELD: Attach @Character2, @Character3, and @Reference Video. Only the two men remain. Character3 rises, gathers physical momentum, and drives one full-body sword thrust into the center of the rectangular shield, never the commander. Sparks and compressed air burst outward; the shield flies sideways while Character3 keeps his sword. Character2 remains standing but exposed. CLIP 13 - 03:38-03:54 - WORLD RELEASE: Attach @Character2, @Character3, and @Reference Video. In one final grounded exchange, Character3 strikes the commander's chest plate without blood. Character2 says in Japanese that this is the end, then dissolves into crimson-gold polygons. The cathedral washes to white. Hard cut to a people-free aerial of an original European floating fortress dissolving into gold-white particles over a sunset cloud ocean. CLIP 14 - 03:54-04:08 - EPILOGUE: Attach @Character1, @Character4, and @Reference Video. In a quiet modern bedroom at dawn, use only their faces in simple contemporary clothes. They sit by an open window and join hands. Finish on clouds clearing over an ordinary sunlit landscape. No headset, fantasy prop, title, or text. GLOBAL NEGATIVE PROMPT: White studio background, character sheet, literal source frame, existing franchise content, copied symbol, readable interface, captions, subtitles, logo, watermark, extra principal, missing principal, duplicate character, identity swap, Character1 and Character3 treated as different people, changed face, changed hair, changed costume, extra sword, missing sword, malformed weapon, round shield, missing shield, Character4 visible during Clip 10, Character4 surviving Clip 11, commander defeated before Clip 13, floating weapon, energy beam, impossible stunt, gore, corpse, text, low resolution, distorted hands, extra fingers, extra limbs, audio mismatch, narration, lyrics, title card. Final Summary Create fourteen consecutive original clips totaling about 4 minutes and 8 seconds. Character1 and Character3 are one hero in beginner and advanced gear; Character4 is the recurring fencer; Character2 is the only crimson commander; Character6 is the supporting lancer. Use the reference video only for broad visual rhythm and motion, while this prompt controls the full original story and continuity. b7:T2325,PROMPT HEADER: Edit the uploaded source video by replacing the two original performers with the two supplied original character references while preserving the source video as the only motion, performance, camera, editing, environment, and audio master. SOURCE VIDEO AND REFERENCE ROLES: Use @your-video as the sole editing master. Do not reinterpret, redesign, restage, extend, shorten, or replace the source performance. Use @character-a only as the global appearance reference for Character A, corresponding to the performer who starts on the left side of the original frame. Use @character-b only as the global appearance reference for Character B, corresponding to the performer who starts on the right side of the original frame. The two PNG reference images are appearance references only. They are not first frames, last frames, keyframes, location references, lighting references, background plates, storyboard panels, or split-screen layouts. Do not show the reference images inside the edited video. CHARACTER MAPPING AND IDENTITY: Replace the original left performer with Character A from @character-a. Preserve Character A’s recognizable skull-style face covering, helmet, dark tactical clothing, equipment layout, colors, materials, and overall silhouette as shown in the reference image. Replace the original right performer with Character B from @character-b. Preserve Character B’s recognizable pale gray-green humanoid appearance, elongated head and body proportions, surface texture, eye design, colors, and overall silhouette as shown in the reference image. Maintain exactly two performers throughout the full video. Keep Character A and Character B as separate identities with stable appearance, clothing or surface structure, masks or facial coverings, helmets or head shapes, colors, materials, proportions, and accessories. Do not exchange identities when they cross the frame. Track each character by the original performer’s motion path, body position, depth relationship, and temporal identity rather than by screen-left or screen-right position after the opening frame. PERFORMANCE PRESERVATION: Preserve the original singing performance from the source video. Match the original performers’ head orientation, torso movement, shoulder and chest rhythm, hand gestures, arm timing, stance, sway, and interaction timing. The replacement characters should appear to perform the same song through body language and the exact original audio timing, without inventing new choreography or extra gestures. Because the characters’ mouths and noses are covered or otherwise not human-readable, do not generate exposed human faces, visible human lips, realistic mouth shapes, or lip movements that conflict with the references. Express singing through head direction, subtle nods, shoulder and chest breathing, body rhythm, hand gestures, posture, and the original performance timing. Keep the replacement motion synchronized to the existing singing audio without generating new dialogue or changing the vocal track. CAMERA, EDITING, AND SPATIAL CONTINUITY: Preserve every original camera cut, shot size, framing, lens perspective, camera movement, zoom, pan, tilt, tracking move, depth of field, focus transition, editing beat, and temporal rhythm. Maintain the original orange photography-studio background, the top-hanging microphones, the performers’ relative depth, their starting left and right positions, their height relationship, the distance between them, and all original stage-space relationships. If the performers briefly cross, overlap, or exchange apparent screen positions, follow each original motion trajectory continuously. Do not swap Character A and Character B simply because one moves to the other side of the frame. Preserve foreground and background order, occlusion, entry and exit timing, and the exact screen-space relationship wherever the source video provides it. LIGHTING, COMPOSITING, AND MATERIAL INTEGRATION: Integrate the replacement characters naturally into the source footage. Match the original orange studio lighting, direction and softness of shadows, highlights, reflections, color temperature, exposure, contrast, perspective, depth of field, grain, compression, and motion blur. Keep character edges stable during movement and camera changes. Make the character materials respond consistently to the source lighting without introducing a new environment, a new color grade, or an unrelated cinematic style. Preserve realistic contact with the original stage space. Keep feet, lower bodies, microphones, shadows, occlusions, and foreground-background relationships aligned with the source video. Do not let reference-image backgrounds, borders, panels, or studio layouts appear in the result. AUDIO AND TIMELINE: Keep the original singing audio exactly aligned to the source video’s timeline, rhythm, dynamics, pauses, and volume changes. Preserve the original music, singing, room sound, microphone presence, and timing only as they exist in the uploaded source. Do not add dialogue, narration, voice-over, subtitles, captions, sound effects, music, or a new vocal performance. Do not retime the audio to fit newly invented movement; fit the character replacement to the original audio and video timing. SAFETY AND CONTENT BOUNDARIES: This is a fictional stage singing performance. Do not depict combat, attack, weapon use, ammunition, explosions, wounds, blood, gore, threats, or real military operations. The performers’ hands may perform only the original stage gestures from the source video. Do not add weapons, dangerous props, political symbols, real military organizations, national flags, brand logos, film logos, title cards, subtitles, captions, or watermarks. ENDING AND OUTPUT STATE: End exactly when the source video ends, with the same final shot, final positions, final gestures, final lighting, final audio state, and final edit timing. The result must be a clean full-screen edited singing video containing exactly two original characters, with no visible reference-image panels, no split-screen character sheet, no extra people, and no generated text. NEGATIVE PROMPT: Changing the source choreography, new choreography, new gestures, new dialogue, narration, voice-over, new music, changed singing audio, changed audio timing, lip-sync animation, exposed human face, visible human lips, human mouth, wrong mask, wrong helmet, wrong head shape, identity swap, face swap between characters, character cloning, extra performer, missing performer, performer duplication, screen-left and screen-right identity exchange, incorrect crossing trajectory, broken occlusion, changed height ratio, changed distance, changed foreground-background order, changed microphone position, missing hanging microphone, new microphone, changed orange studio background, reference-image background, white reference sheet, dark alien environment, battlefield, weapons, ammunition, combat, attack, explosion, wounds, blood, gore, threat, realistic military operation, political symbols, national flags, brand logo, film logo, subtitles, captions, title cards, watermarks, split screen, triptych, character turnaround sheet, model sheet, storyboard panels, image borders, extra props, dangerous hand actions, deformed hands, extra limbs, unstable clothing, changing armor, changing mask, changing helmet, changing colors, material flicker, texture flicker, facial drift, anatomy drift, temporal flicker, frame interpolation artifacts, frozen body, stiff motion, broken tracking, floating feet, incorrect shadows, wrong perspective, mismatched depth of field, missing motion blur, excessive motion blur, low quality, blurry character replacement. RIGHTS AND SOURCE-MATERIAL NOTE: Use the source video and its music, singing, and imagery only when you own them or have permission to edit them. If the source contains third-party material and you do not have authorization, replace it with a self-owned or properly licensed singing video and audio before using this workflow. Final Summary Use the original singing video as the sole editing master and replace only the two performers. Character A from image 1 takes the original left performer’s identity track, while Character B from image 2 takes the original right performer’s identity track. Preserve the original singing performance, audio timeline, camera cuts, framing, orange studio background, hanging microphones, lighting, shadows, perspective, depth of field, motion blur, and spatial relationships. The two reference images control appearance only. They must never become first frames, end frames, backgrounds, split-screen panels, or character sheets inside the output. Track each identity through crossings and camera changes using the original motion path, not the character’s temporary screen position. The finished result is a two-character fictional stage performance with no added dialogue, subtitles, weapons, violence, or unrelated visual elements. b8:T1f0d,PROMPT HEADER: REFERENCE ROLE: Use the uploaded character reference as the main identity guide. Preserve the character’s facial identity, body shape, hairstyle, clothing style, posture, and overall presence. The exact same reference identity appears as two versions of one person: one version is hanging from the edge of a dark sea cliff, and the other version is standing above as the rescuer. Both versions must share the same face, proportions, skin tone, hairstyle, gaze quality, and overall presence. Do not create two different identities. FORMAT AND VISUAL STYLE: Create a 15-second vertical 9:16 cinematic video. Set the story on a stormy ocean cliff with black wet volcanic rocks, crashing blue-green waves, cold mist, sea spray, a gray overcast sky, and dramatic coastal wind. Use a cinematic, emotional, fashion-editorial, realistic high-end short-film style with shallow depth of field, soft cold daylight, a muted teal-gray palette, tactile water droplets, and physically believable movement. WARDROBE: The cliff-edge version wears the exact same outfit shown in the character reference image. Preserve the reference outfit’s silhouette, fabric texture, colors, layering, fit, footwear, and overall styling. The rescuer version wears a refined white suit set with a clean structured blazer, a simple white inner layer, white tailored trousers, and a white wide-brim hat. Use minimal neutral tailoring. Keep both outfits elegant and weather-worn by mist and sea spray. Preserve wardrobe continuity within each version across all shots. STORY AND EMOTIONAL FUNCTION: A single identity confronts danger and rescue at the same time. One version is nearly falling from the cliff and grips the rock with both hands. The other version reaches down calmly and pulls the cliff-edge version back to safety. The emotional meaning is self-rescue, inner strength, survival, quiet confidence, and the realization that help can come from within. SHOT TIMELINE: 0.0–2.0 SECONDS — CLIFF HOLD: Use a high-angle overhead shot. The cliff-edge version clings to the black rock ledge with both hands while violent ocean foam churns far below. The body is pressed against the cliff edge, making small desperate attempts to climb upward. Each effort lifts the body only slightly before gravity pulls it back down. The head tilts upward toward the top of the cliff with fear and effort. Show wet hair, soaked clothing, trembling fingers, unstable breathing, and realistic body weight. The camera slowly pushes downward, creating controlled vertigo without losing the character’s anatomy. 2.0–4.5 SECONDS — WAVE IMPACT: Continue the high-angle overhead perspective. A heavy wave crashes upward against the cliff. Sea spray washes across the lens and partially covers the frame. The cliff-edge version keeps the head angled upward while gripping the ledge tightly. The character continues trying to climb, but the wave and gravity force the body back against the cliff. The powerful wave impact peaks during this section. 4.5–5.6 SECONDS — EXHAUSTED FACE: Use a close shot from ledge level. The face rises into frame, wet, exhausted, and frightened, with the eyes looking upward. Fingers scrape against the dark stone. Keep the background ocean blurred but visibly violent and moving. 5.6–7.1 SECONDS — REACHING HAND: A hand enters from the upper frame and reaches downward. Shift focus from the exhausted face to the approaching hand. The cliff-edge version hesitates for one brief beat, then reaches toward the rescuer’s hand. Keep the hesitation visible before contact. 7.1–8.4 SECONDS — RESCUER REVEAL: Use a reverse angle to reveal the rescuer version above the ledge, wearing the white wide-brim hat and pearl-textured white outfit. The rescuer kneels low with one arm extended, calm and focused. Mist blows across the frame. Hold a low heroic angle from below the ledge. The rescuer’s face remains visible and readable. 8.4–10.5 SECONDS — HAND CONTACT: Use an extreme close-up of the two hands. Fingers almost touch, slip slightly because of water, and then lock firmly together. Keep the hand contact realistic: one clear grasp, no extra fingers, no duplicated hands, no broken anatomy, and no instant teleportation into a locked grip. 10.5–13.2 SECONDS — PULL-UP: Use a wider shot. The rescuer version pulls with steady force while the cliff-edge version climbs over the ledge. One knee rises onto the rock. Boots scrape against the wet cliff. Ocean mist surges behind them. The camera shakes subtly with the physical effort, but the movement remains controlled and readable. 13.2–15.0 SECONDS — CALM SELF-BELIEF: Use a tight emotional close-up after the rescue. The rescuer version holds the cliff-edge version close. The rescuer’s right hand gently strokes the back of the cliff-edge version’s head in a comforting gesture. Keep the rescuer’s entire face clearly visible and well lit. The wide-brim hat must not hide the face or obscure the eyes; angle the hat slightly upward or position the camera so both eyes remain fully readable. The rescuer looks directly into the camera with a gentle yet determined expression, conveying self-belief, quiet confidence, emotional control, and a relaxed sense of ease. End on this calm, self-assured gaze with soft sea mist and muted coastal light. Do not turn the face away, hide it under the hat, or cover the eyes with shadow. ON-SCREEN TEXT: Only one subtitle appears at the lower center throughout the video in elegant white serif type with a subtle shadow: “Only you can save yourself No one else...” No other readable text appears. AUDIO: Begin with low cinematic ambience, distant wind, deep ocean rumble, and tense soft strings. Add a strong crashing-wave sound at 3.5–4.0 seconds. Keep breathing and wet-hand contact subtle but audible during the close-ups. Add a rising emotional swell from 8.4 seconds as the hands connect. Peak the music and ocean impact around 11.0–13.0 seconds during the pull-up. End with softer waves, wind, and a warm resolved musical tone. CONTINUITY RULES: Both visible versions are the exact same reference identity, styled differently only by situation and wardrobe. Preserve the face, body shape, hairstyle, and overall presence across every shot. Maintain the same cliff location, wet black-rock texture, stormy ocean direction, mist behavior, subtitle placement, and emotional arc. Do not add random extra people. Do not allow identity drift, glamour-pose breaks during the danger sequence, comedy, fantasy glow, superhero effects, a clean studio background, extra captions, distorted hands, duplicated limbs, or changing outfits mid-shot. NEGATIVE PROMPT: two unrelated identities, changed face, face drift, changed body shape, changed hairstyle, changed skin tone, extra people, extra rescuers, extra cliff-edge characters, duplicated limbs, extra fingers, fused fingers, broken hand contact, missing hands, impossible climbing, teleportation, instant grip, falling to the ocean, death, blood, injury, fantasy glow, superhero effects, clean studio background, glamour pose during danger, changing wardrobe, hidden rescuer face, hat covering the eyes, heavy eye shadow, camera losing the subject, random camera cuts, excessive camera shake, subtitles other than the exact provided line, extra text, logos, watermark, cartoon, anime, glossy CGI, low resolution Final Summary Create a 15-second vertical cinematic video in which one character confronts danger and rescues herself. Use one uploaded character reference to preserve the same face, body shape, hairstyle, gaze quality, and overall presence across both visible versions.The sequence moves from a dangerous cliff hold to a close hand connection, a controlled pull-up, and a calm final embrace. The stormy ocean, wet volcanic rocks, cold mist, wardrobe contrast, subtitle, and audio arc work together to express self-rescue, inner strength, and survival. b9:T1bac,【参考素材定义】图片1＝戴夫唯一的人物身份参考：图片2＝唯一感染向日葵的形态参考：图片3＝四只不同僵尸的身份参考：图片4＝唯一一台红色旧割草机的道具参考：图片5＝唯一末日庭院的场景参考：【视频要求】生成一条从0秒开始的30秒、16:9照片级写实电影视频，完整讲述戴夫从房前发现向日葵受袭、放弃启动割草机、拿铲赶去救它、挖根失败、险些被感染的花咬到，最后被花推回安全步道的连续事件。不是两条视频拼接；第8秒只延续同一个人的动作，不重新开场、不重置空间或道具。全片自然速度、明确物理接触。割草机在这30秒内始终熄火停在房前；戴夫真正启动机器留给下一条视频。声音只有庭院风声、僵尸低吼与啃咬、启动绳轻响、脚步、铲面撞击与铲土、黑根绷紧、花盘咬合、戴夫呼吸，以及三句英文对白“Hold on!”、“Come on…”、“No!”。不添加背景音乐、配乐或旋律性音效。【场景环境】画面始终遵守左房右入口的空间轴线：左侧房屋与门廊 → 戴夫及门前割草机 → 中段土坑与向日葵、两只近处僵尸 → 右后方破栅栏与两只远处僵尸。割草机四轮落在房前草坪的清场起点，机身前部与花相隔约三米空草坪；门廊外没有第二台机器。木柄铲靠在戴夫右后方一步可及的门廊立柱上。南侧石步道沿草坪前缘从房子通向花，是戴夫救援和撤退的唯一通路。戴夫从画面左侧沿步道向右赶去；到花的南偏西侧后，用铲子把瘦高僵尸向草坪北侧推开，不把僵尸推向摄影机。花在戴夫右前方，铁桶僵尸在花后侧，路障与歪领带僵尸始终更靠右。结尾花的叶片把戴夫向画面前方的南侧步道推开；它的黑根只困住两只近处僵尸，另外两只尚在靠近。【摄影机设置】摄影机起始位置：草坪南侧、房前石步道外缘，位于面向花的戴夫右侧后方约四米处，整条视频都留在房屋—割草机—向日葵轴线的南侧。拍摄距离：开场用全景同框呈现门廊、戴夫全身、完整割草机、花和四只僵尸；跟拍救援时距戴夫约三米；根部和花盘近景距主体约一至两米。机位高度：开场约1.7米微高斜侧视角，跟拍约1.3米，花盘近景约1.2米；结尾全景约1.4米。拍摄方向：从南向北偏东看，画面左侧一直是房屋与机器，右侧是花与栅栏；戴夫的前进方向始终是画面左向右。镜头焦段：开场28毫米全景；阳光50毫米近景；房前拿铲35毫米中全景；沿步道跟拍35毫米；挖根50毫米近景；花盘70毫米近景；结尾35毫米同侧全景。运镜方式：开场沿南侧向右轻移半米，露出完整机器前轮和它正对的花；第3秒切花叶近景轻推；第5秒切回房前同侧；第8秒沿南侧步道向右平行跟拍戴夫，不返回房前重新开始；第12秒切到土坑同侧近景；第16秒切花盘近景；第19秒切回同侧全景。每次切镜均保持左房右花，不反打、不绕轴。【动作时间轴】【0—5秒】房前微高斜侧全景立即建立全部站位。左侧可见房屋墙面与门廊，戴夫双脚站在门廊前地面步道，左手扶割草机后推把，右手搭在发动机启动绳把手上但没有拉。完整机身与四轮位于他前方的草坪起点，前轮朝画面右侧的向日葵。花在中段土坑，被瘦高与铁桶僵尸从两侧啃咬；路障与歪领带僵尸从更右方走来。镜头向右轻移半米，仍让戴夫、机器、花和僵尸同框；机器不动、引擎无声。【5—8秒】同轴南侧切到花叶近景。花忍着两侧啃咬，根仍固定，一片主叶向画面左侧房子与戴夫的方向伸出，叶上托着唯一阳光。阳光短促亮起，暖光先照叶缘再照地面；戴夫画外呼吸停住，僵尸没有换位。【8—13秒】切回房前同侧中全景，同时保留门廊、完整机器和远处花。戴夫眼睛先锁定阳光，随后转头；右手松开启动绳，绳把回到发动机，左手放开后推把。他右手从身后门廊立柱旁拿起唯一木柄铲，沿机器南侧的石步道朝画面右侧花的位置迈出第一步，急声说英文：“Hold on!” 末帧右手持铲、身体和脚尖向右，机器熄火留在原位。【13—18秒】动作直接接上一个迈步。摄影机在南侧以35毫米平行跟拍，戴夫沿同一石步道向右跑过机器正面与草坪前缘，到中部花的南偏西侧；房屋与割草机逐渐留在他身后画面左侧，不能让他在花旁凭空出现。他用右手的铲面把挡路的瘦高僵尸向北侧草坪推开，僵尸脚落地踉跄退开，并未被击飞。戴夫顺势把铲刃插在花根旁的土中，左脚踩稳步道边缘。【18—22秒】切土坑同侧近景并保持戴夫和花的相对位置。戴夫右手下压铲柄，左脚抵住松土，真实土块翻起，但地下黑根绷得更紧；他右手始终握铲柄，左手托住花茎向上试抬，根部纹丝不动。戴夫喘着说英文：“Come on…” 阳光仍在另一片叶上，没有跳到他手里。先前被推开的瘦高僵尸在后景重新站稳。【22—26秒】切花盘近景，画面一侧仍看见戴夫的左手和握铲的右手。感染抽动使花突然向他张口，戴夫左手立即从茎上缩回；花在咬到他之前强行转开花盘，咬向空处。戴夫右手没有放掉铲柄，脚仍在土坑南侧；后景瘦高僵尸开始重新逼近。【26—30秒】切回轴线南侧全景。戴夫仍弯身靠花，花的一片主叶真正抵住他的左肩，向南侧安全步道推他。受力后戴夫先把铲刃从土里拔出，右手握铲，脚从土坑边先后退到前景步道，喊英文：“No!” 花的黑根从固定土坑向两侧伸出，先缠住重新靠近的瘦高僵尸脚踝，再缠住铁桶僵尸脚踝；两只僵尸因真实拉力停步挣扎。路障与歪领带僵尸还在右侧接近，未提前受困。末帧戴夫在南侧步道、右手持铲，花仍扎根，唯一阳光仍在叶上，房前割草机仍熄火未动。所有动作在同一片庭院连续发生；第8秒没有新开场或姿势重置。手中铲子、绳把、阳光与两只近处僵尸的位置每镜承接前镜。【反应顺序】戴夫先看到花送来的光，眼睛改变后才转头、松绳、拿铲。挖根时先感到铲柄与根部阻力才加力；花突然张口距离过近时，他的左手因近身威胁立即缩回；叶片实际抵到左肩后，肩和双脚立即按推力退开，不人为延迟。【节奏要求】前5秒让观众同时理解割草机将清掉这一排、向日葵正在被啃咬、它却仍给戴夫阳光。8—12秒完整看见戴夫从房前跑到花旁，不能省略空间移动。救援、失控、保护按自然速度连续推进；不添加额外战斗、慢动作或长时间停顿。ba:Tf9d,Part-1 | 30s | 4:3 | 24fps | Photorealistic Live-Action Luxury Commercial Recreate the reference advertisement as closely as possible in shot structure, timing, framing, lighting, camera movement, macro photography, transitions, editing rhythm, and overall cinematic language. LOCKED ELEMENTS: • Same Black male model (~30, athletic build, black tank top, black trousers) • Same AUREL VX-01 watch (carbon tonneau case, skeleton dial, silver hands, red crown, black strap) • Same BLACK VANTA R1 supercar • No changes to face, watch, car, proportions, materials, or colors VISUAL STYLE: Pure black environment, deep charcoal shadows, metallic silver highlights, dark crimson accents, luxury motorsport atmosphere. Real cameras, real lenses, real lighting, real reflections, real depth of field, realistic motion blur. No CGI look, no render look, no AI artifacts, no morphing, no floating objects, no deformed hands, no plastic skin. SEQUENCE: Mechanical assembly macro → shadowed model portrait → red/black engineering macros → rotating gears → suspended watch beauty shot → crown/strap/movement macros → dark crimson particle transition → model checks watch → VANTA R1 reveal → speed transition → wheel macro → watch hero macros → frontal watch product shot → model walking with car → hero pose beside car → rainy lineup of multiple VANTA R1 supercars. AUDIO: Dark luxury industrial-electronic soundtrack, deep sub-bass, subtle ticking, metallic clicks, restrained engine rumble, cinematic impacts. No dialogue. No voiceover. EDITING: Elegant luxury-commercial pacing, hard cuts synchronized with impacts, minimal motion-blur transitions. No text, logos, captions, or typography. OUTPUT: Ultra-photorealistic premium cinema commercial. Natural skin, real materials, realistic reflections and shadows, consistent character, watch, vehicle, and lighting. ZERO AI VIBE. PART 2 — DIRECT CONTINUATION 16.8s | 4:3 | 24fps | Photorealistic Live-Action CONTINUITY LOCK: Continue directly from Part 1's final rainy BLACK VANTA R1 formation. Keep the exact same male model, face, skin tone, hair, physique, wardrobe, AUREL VX-01 watch, watch movement, red crown, strap, BLACK VANTA R1, burgundy interior, lighting, color grade, lens behavior, film grain, reflections, motion blur, and cinematic style. No redesigns. No character drift. No watch changes. No car changes. No CGI look. No AI artifacts. No morphing. No plastic skin. SEQUENCE: • Extreme macro tracking along the AUREL VX-01 carbon case edge with visible screws and red crown highlights. • Abstract black-and-silver metallic surface macro resembling luxury automotive reflections. • Inside BLACK VANTA R1: same driver seated behind the wheel, watch visible, burgundy interior, controlled lighting. • Wrist close-up on steering wheel, watch catching dashboard light. • Extreme macro of the skeletonized movement, gears, bridges, screws, and subtle mechanical motion. • Hero watch shot against deep crimson-black studio background, slow push-in. • Return to car interior; camera tracks outward through the side window, revealing more of the vehicle while keeping the watch visible. • Final hero product reveal: slow rotation/dolly around AUREL VX-01 into a premium three-quarter hero angle, then hold a nearly static luxury product shot until the end. AUDIO: Continue the exact same industrial-electronic score from Part 1. Deep sub-bass, subtle ticking, metallic details, restrained engine rumble during car shots only. No restart, no dialogue, no voiceover. EDITING: Elegant luxury-commercial pacing, clean hard cuts, realistic motion blur, no text, logos, taglines, or graphic overlays. OUTPUT: Premium luxury-watch commercial filmed with real cinema cameras, macro lenses, real materials, natural skin, realistic carbon fiber, sapphire crystal, reflections, shadows, and optical depth of field. Consistent character, watch, vehicle, and lighting throughout. ZERO AI VIBE. bb:T1b66,Part-1 30-SEC VERTICAL 9:16 | PHOTOREALISTIC LUXURY BEAUTY COMMERCIAL Create a premium European luxury beauty commercial using the provided CHARACTER REFERENCE and PRODUCT REFERENCE. ABSOLUTE REALISM: Live-action cinema quality. Real human model, practical cosmetics, real glass, real liquids, real fruit and physical studio sets. Natural skin pores, realistic eyes, hair strands, hands, reflections, liquid physics and camera optics. No CGI look, wax skin, uncanny eyes, plastic face or AI aesthetic. CHARACTER LOCK: Use the exact same blonde female model throughout. Preserve her face, blue-green eyes, high bun, makeup, body proportions and dusty-rose pink suit. PRODUCT LOCK: Use the exact VELVET KISS raspberry-red lip oil in every product shot. Preserve bottle shape, color, lettering, glass and rose-gold cap. SHOT TIMELINE: 0–1.7s — Extreme macro product reveal inside glossy raspberry-red liquid, droplets, rose-gold highlight, slow push-in and subtle rotation. 1.7–4.1s — Model walks through a pristine white luxury studio, then approaches camera holding VELVET KISS close to lens. Natural focus shift from product to eyes. 4.1–7.3s — Realistic hands open the lip oil. Macro applicator movement and close-up lip application with natural glossy texture. 7.3–10.6s — Product hero surrounded by raspberry-red liquid, followed by glossy lips and a chrome-surface product shot. 10.6–13.5s — Product inside soft blush-pink liquid folds. Realistic fluid physics, macro details and slow camera push toward VELVET KISS branding. 13.5–18.8s — Model in dusty-rose suit holds product beside her cheek, applies it again, then looks into camera with subtle natural expressions and blinking. 18.8–22.4s — Editorial upside-down portrait, followed by floating cosmetic droplets and a transition into realistic fresh raspberries and strawberries. 22.4–26.6s — Product among fresh berries, then model returns to the white studio. She walks toward camera, leans in and holds the product near her lips. Final clean product hero on dusty-rose background. 26.6–30s — Brief realistic blue-toned editorial beauty contrast, then immediately return to the model in white studio holding VELVET KISS beside her face. Subtle final camera push-in. AUDIO: Sophisticated minimal electronic-pop luxury soundtrack with deep bass pulse, glossy synth textures, subtle sparkle and restrained percussion. No vocals or dialogue. Add faint realistic cap-twist, applicator and liquid sounds. EDITING: Premium fashion-commercial pacing. Precise music-synced cuts, elegant slow motion, fast editorial transitions and realistic focus pulls. No AI morphing, text, subtitles, promotional graphics or watermarks. CAMERA: Full-frame cinema look with 35mm, 50mm, 85mm and 100mm macro lenses. Realistic depth of field, motion blur, exposure, reflections and controlled studio lighting. FINAL QUALITY: Photorealistic live-action luxury beauty campaign. Maintain perfect character and product continuity in every frame. Same face, hair, makeup, outfit, body, product design, color, proportions, lettering and cap throughout. Every shot must feel like it was filmed during one professional production. Part-2 + Refrences Character+Storyboard+Product CONTINUE DIRECTLY FROM THE FINAL FRAME OF PART 1. CRITICAL CONTINUITY LOCK: The first frame of this generation must match the final frame of Part 1 exactly in character identity, face, hairstyle, makeup, clothing, body position, product position, product design, studio environment, lighting direction, lens characteristics and overall color grade. SAME CHARACTER: 27-year-old blonde female fashion model, fair realistic skin, blue-green eyes, high ballerina bun, defined cheekbones, natural glossy raspberry-red lips, dusty-rose pink tailored suit and matching trousers. SAME PRODUCT: VELVET KISS luxury raspberry-red lip oil, clear glass cylindrical bottle, translucent deep raspberry-red liquid, polished metallic rose-gold cap, same dimensions, same branding and same applicator. STYLE: Photorealistic live-action luxury beauty campaign. No AI appearance. No character drift. No product redesign. No fantasy CGI. No InVideo branding or watermark. 30.00–31.10 Continue with the same model standing in the pristine white studio, holding the VELVET KISS lip oil beside her face. She subtly turns her head toward camera. Natural blink and tiny facial movement. Maintain exact pose continuity from the final frame of Part 1. 31.10–32.20 Cut to an extreme macro product shot. VELVET KISS rests in a thin layer of glossy raspberry-red liquid. Camera slowly pushes closer as realistic reflections move across the glass and rose-gold cap. 32.20–33.30 Macro beauty shot of the same product surrounded by silky translucent raspberry-red liquid. The liquid folds naturally around the bottle with physically believable movement. No artificial CGI appearance. 33.30–34.30 Return to the model in the white studio. Medium close-up. She holds the product beside her cheek, looking directly into camera. She gives a very subtle natural smile, keeping the sophisticated luxury editorial mood. 34.30–35.30 Close-up of the model's lips and lower face. The glossy raspberry-red lip oil catches the studio light. She slightly turns her head, creating a natural highlight across the lips. 35.30–36.60 Final hero shot: VELVET KISS product lying diagonally inside a smooth glossy raspberry-red liquid surface. Slow cinematic camera slide from left to right. Product remains perfectly sharp and centered enough to clearly recognize the packaging. 36.60–37.75 Clean final product beauty frame. VELVET KISS stands upright against a sophisticated dusty-rose/pale-pink studio background with a soft realistic shadow beneath it. The rose-gold cap catches one elegant highlight. Camera performs an extremely subtle push-in and then settles completely. END EXACTLY ON THE CLEAN PRODUCT HERO FRAME. No fade into black. No end card. No InVideo logo. No watermark. No extra text. No artificial UI. No sudden camera movement. No additional scene after the final product frame. AUDIO CONTINUITY: Continue the exact same luxury electronic beauty soundtrack from Part 1 without restarting the music. The final 2 seconds become slightly quieter and more refined, leaving a clean premium ending. Keep subtle cosmetic handling and liquid ambience underneath. No dialogue, no narration, no vocals. CAMERA AND IMAGE QUALITY: Same professional full-frame cinema camera aesthetic as Part 1. Realistic macro optics, natural depth of field, physically correct reflections, realistic glass refraction, natural skin texture, realistic liquid physics, controlled studio lighting, natural motion blur. FINAL REQUIREMENT: The entire 37.75-second commercial must feel like ONE continuous real-world production filmed with the SAME model, SAME product, SAME camera package, SAME lighting team, SAME set and SAME color grade. No visual discontinuity between Part 1 and Part 2. bc:Tcf9,SCENE CONTEXT Premium 30-second Japanese cherry juice commercial: playful, surreal, fashion-forward, photorealistic and highly polished. Energetic pacing, Japanese fashion-ad aesthetics, creative HARD CUTS, MATCH CUTS and WHIP CUTS. ACTIVE REFERENCE <<>> is the exact SIPPO Cherry carton reference. Preserve its matte powder-pink rectangular carton, folded top, white flexible straw, striped sunset symbol, dark-red SIPPO logo, Cherry label and two smiling cherries. Never change the packaging into another container. CHARACTERS Four adult Japanese women, 22–28, distinct hairstyles: black bob, raven ponytail with burgundy ribbons, copper-brown bob, waist-length black hair with blunt bangs. Tokyo fashion styling in cherry red, pink, ivory and chrome. Stable faces, natural skin and believable friendship. FIRST FRAME Carton already dominates the foreground with its full front panel readable; lead woman directly behind it. No establishing shot. FORMAT MODE Fast multi-shot commercial with consistent product, characters, wardrobe and color world. Every shot introduces a new visual idea. ACTION AND CAMERA SEQUENCE OPENING — 107° wide rectilinear shot. Carton fills foreground on glossy pink table. Lead woman smiles and points at straw as camera rapidly pushes in. WHIP CUT — Camera bursts through a cherry-red circular portal into a surreal Tokyo studio street with pink tiles, cream walls, chrome rails and giant cherries. Four women walk toward camera in synchronized formation. 84° wide stabilized dolly. MATCH CUT — Carton passes through a circular wall opening and emerges into another woman’s hands inside a burgundy/pink Japanese photo booth. Fast 180° orbit as she sips and side-eyes camera. OVERHEAD — Four women form a radial composition around a cherry-red platform, passing the carton clockwise while camera rotates opposite. Two cherries roll across frame as transition. MACRO INSERTS — Condensation, flexible straw, fingertips, glossy cherry, SIPPO logo and smiling cherry graphics. Product remains faithful to <<>>. LOW ANGLE — Four women surround the product pedestal. Camera rapidly cranes upward as they turn toward camera and break into natural laughter. Realistic hair and fabric motion. FINAL GROUP — 84° wide cherry-red stage with giant circular sunset-inspired light. Each woman holds an identical carton and they perform a synchronized toast. FINAL PACKSHOT — HARD CUT to cream-pink studio. Exact SIPPO Cherry carton on polished pedestal with two cherries beside it. Straw bends right. Slow premium telephoto push. Hold for brand recognition. LIGHTING AND IMAGE QUALITY Premium Japanese beauty-commercial lighting, soft overhead light, cherry-red edge light, pearlescent skin, powder pink/burgundy/cream palette. Photorealistic ARRI Alexa 35 look, natural grain, HDR, pristine commercial finish. PHYSICS Realistic carton weight and stiffness, natural finger pressure, flexible straw, gravity-driven condensation, believable hair, fabric and cherry motion. AUDIO Original upbeat Japanese electro-pop with female vocal chops, bass, claps and sparkling synths. Add synchronized carton taps, straw flex, cherry rolls, fabric swishes, whooshes and sip sounds. Final toast: “SIPPO Cherry!” End with a two-note sonic logo. No subtitles. bd:Tc6e,REFERENCE & SUBJECT Use "@ 是 image1 角色卡上的成年女性：棕色低发髻、脸侧碎发、白色长袖裹身短上衣、浅灰米色运动内搭、灰色高腰短裤。作者没有公开这张角色卡，需要自己先做一张。"},{"number":3,"title":"第三步：写主体定义和强制结构约束","description":"subject_definitions 把 Video1 定成鞋子、image1 定成人物、Audio1 定成白模原音轨；再用【强制结构约束】写死：足弓下方必须左右贯通、能看到背景，不能被鞋底或鞋面填实，活动件的轨迹、幅度、速度都照参考视频。"},{"number":4,"title":"第四步：按时间码写 5 个镜头，一次生成 15 秒","description":"Shot 1 全身起身 → 00:02.5 鞋子侧面低机位特写 → 00:05.5 中景推脸忍笑 → 00:08.5 迈步怕痒、坐回长凳 → 00:10.5 脱右鞋挠脚底。声音只参考 Audio1 原有的层，加轻笑和吸气，不加台词。作者在评论区回复用的是 Seedance 2.5。"}],"references_detail":[],"storyboard":[{"number":1,"description":"0–2.8s 全身：她穿着灰色尖刺高跟鞋坐在长凳上，低头看鞋后扶凳站起。"},{"number":2,"description":"2.8–5.75s 鞋子侧面低机位特写：镂空鞋弓里的弧形活动件来回拨动脚底。"},{"number":3,"description":"5.75–8.9s 中景坐姿，镜头推到脸部特写：憋笑、闭眼、再笑开。"},{"number":4,"description":"8.9–15s 全景：站着笑到弯腰，坐回长凳，脱下一只鞋放在旁边，捧着脚笑。"}],"constraints":"鞋子外形、镂空结构和活动件运动全照 Video1，镂空不能被鞋底或鞋面填满；人物照 image1 角色卡，全片同一人同一双鞋；声音只参考 Audio1 原有的层，不加台词和配乐。与成片不符：提示词写 9:16，成片是 834×1112（3:4）；提示词分 5 个镜头，成片 8.9s 后是一个连续全景，没有单独切「脱鞋挠脚」镜头，也看不清手指挠的是不是足弓；白模里的活动件更宽、像舌头，成片里是细一些的弧形钩。缺口：作者没有公开 image1 角色卡；提示词是成片发出约 3 小时 44 分后才补在另一条帖子下面的，作者没有明说这就是这条成片用的那版；白模的制作方法没有公开。","video_prompt":{"title":"挠痒高跟鞋试穿 · 白模视频 + 角色卡 · 15s","subtitle":"Seedance 2.5 · 中文完整提示词（作者自回复长帖；开头的站点推广链接未收录）","content":"$2f"}},{"id":"zephyra-tiny-gimbal-camera-unbox","title":"口袋云台相机开箱 · 创作者向评测","subtitle":"X · @ZephyraLeigh · Seedance 2.5 · 30秒 · 16:9","description":"Seedance 2.5 口袋云台相机开箱：女博主 Zephyra 拆箱、比大小、讲三轴防抖，30 秒一段生成带口播。","video":"https://media.liuyidaoai.com/tutorials/zephyra-tiny-gimbal-camera-unbox/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/zephyra-tiny-gimbal-camera-unbox/poster.jpg","duration":"30秒","durationSec":30,"styleLabel":"科技开箱","shots":6,"references":0,"model":"Seedance 2.5","style":"YouTube 科技评测 · 创作者工作室 · 口播 + 产品微距","aspectRatio":"16/9","sourceUrl":"https://x.com/ZephyraLeigh/status/2103849346198655342","sourceAuthor":"@ZephyraLeigh","sourcePlatform":"X","sourceImpressions":943,"sourceStats":{"asOf":"2026-09-27","likes":22,"reposts":3,"bookmarks":6},"formats":["产品广告"],"hook":{"structure":"口播开场 → 俯拍开箱 → 比大小 → 云台演示 → POV 试拍 → 第一印象","opening":"第 0 秒是双手按着桌上还没拆的相机包装盒，约 2s 切到 Zephyra 拿着盒子对镜头开口。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"约 5s 俯拍掀开盒盖，从盒里取出相机；约 8s 回到桌前中景摆放配件；约 11s 把相机举在手机旁边比大小。","at":5},{"title":"功能演示","text":"约 13–16s 手部近景转动相机、露出云台和屏幕；约 17s 她站起来拿着相机走几步。","at":13},{"title":"结尾怎么收","text":"约 21–23s 切成相机视角在工作室里移动的试拍画面；约 24s 回到她把相机举近镜头，说出第一印象结束。","at":21}],"copyThis":"提示词按秒写死 6 段（钩子、开箱、盒内物品、功能、实测、第一印象），每段一个机位加一句台词，口播和产品镜头交替切。","approx":true},"tags":["30秒 · 科技开箱","16:9 横屏","Seedance 2.5","口播 + 产品微距","同一虚拟博主系列"],"steps":[{"number":1,"title":"第一步：锁定博主形象和工作室","description":"提示词开头要求保留 Zephyra Leigh 的脸、发型、身材、声音和人设，并给她换一身固定服装：米色针织上衣、深棕色高腰裤、金色手表、小耳饰和细手链；场景是摆着摄像机、麦克风、笔记本和柔光灯的创作者工作室。作者没有公开人物参考图。"},{"number":2,"title":"第二步：按秒写 6 段口播和机位","description":"0–4s 钩子（相机从盒中取出、凑近镜头、自我介绍）→ 4–9s 俯拍开箱 → 9–14s 摆出盒内物品并和手机比大小 → 14–21s 云台微距和触屏演示 → 21–26s POV 式试拍 → 26–30s 第一印象。每段都写好了英文台词。"},{"number":3,"title":"第三步：整段贴进 Seedance 2.5","description":"把下方完整提示词贴进 Seedance 2.5，一次生成 30 秒带口播的视频。结尾写了镜头、声音、表演、画质要求和禁止项：全名只在开场说一次，服装保持一致，不要科幻元素、夸张反应、畸形手指和假的产品互动。"}],"references_detail":[],"storyboard":[{"number":1,"description":"0–5s 双手按着未拆的相机包装盒；切到 Zephyra 拿着盒子对镜头自我介绍。"},{"number":2,"description":"5–8s 俯拍掀开盒盖，取出相机，盒里有相机、保护套、配件。"},{"number":3,"description":"8–12s 回到桌前中景摆放配件，把相机举在手机旁边比大小。"},{"number":4,"description":"13–19s 手部近景转动相机、露出云台和屏幕；她站起来拿着相机走几步。"},{"number":5,"description":"20–23s 相机装在杆上，切成相机视角在工作室里移动的试拍画面。"},{"number":6,"description":"24–30s 回到她把相机举近镜头，说出第一印象，自然微笑。"}],"constraints":"保留同一位博主的脸、发型、身材、声音和人设；服装全程一致；全名只在开场说一次；真实的手部、手指和产品互动，不要科幻、全息、卡通、夸张反应和画面故障。与成片不符：开场不是提示词写的「从盒中取出相机的微距」，而是先拍未拆的包装盒；包装盒上的品牌字是乱码（类似「NUCEPOVIG」）；21–26s 的 POV 试拍拍的是空工作室，没有像提示词写的那样跟拍她的脸。缺口：作者没有公开人物参考图；口播音频没有转录核对。","video_prompt":{"title":"Tiny Gimbal Camera Unboxing · Creator Review · 30s","subtitle":"Seedance 2.5 · 16:9 · 英文完整提示词（作者自回复长帖）","content":"$30"}},{"id":"eej-ooxo-selfie-hand-sweep-face-switch","title":"手一扫就变脸 · 双人自拍整活","subtitle":"X · @EEJ_OOXO · 3 张参考图 + 动作参考视频 · 10秒 · 1:1","description":"动作参考视频驱动的双人自拍整活：闺蜜手一挡一扫，前面的她在咧嘴笑和面无表情间来回切换，最后两人笑崩。","video":"https://media.liuyidaoai.com/tutorials/eej-ooxo-selfie-hand-sweep-face-switch/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/eej-ooxo-selfie-hand-sweep-face-switch/poster.jpg","duration":"10秒","durationSec":10,"styleLabel":"自拍整活","shots":1,"references":0,"model":"未注明（参考图 + 动作参考视频驱动）","style":"手机前置自拍 · 粉色阁楼卧室 · 一镜到底","aspectRatio":"1/1","sourceUrl":"https://x.com/EEJ_OOXO/status/2103905933999091826","sourceAuthor":"@EEJ_OOXO","sourcePlatform":"X","sourceImpressions":35597,"sourceStats":{"asOf":"2026-09-27","likes":280,"reposts":34,"bookmarks":277},"formats":["角色表演","手机POV·Vlog"],"hook":{"structure":"两人板着脸 → 手扫过就变脸，反复切换 → 一起笑崩","opening":"第 0 秒是手机前置自拍：戴橙色星星帽、扎双麻花辫的女孩在前，黑长发穿灰西装的闺蜜贴在她右后方，两人都面无表情看镜头。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"约 0.5s 起，后面的闺蜜两只手一上一下交替从前面女孩脸上扫过；每扫一下，前面的她就在咧嘴大笑和闭嘴面无表情之间切换一次，闺蜜全程板着脸，一直持续到约 8.5s。","at":0.5},{"title":"结尾怎么收","text":"约 9s 闺蜜放下手，两人同时笑崩，张大嘴、眯眼、肩膀抖动，笑着结束。","at":9}],"copyThis":"动作和表情节奏全部交给一段真人动作参考视频，提示词只负责换人、换场景和补最后 1 秒的笑场。","approx":true},"tags":["10秒 · 自拍整活","1:1 方屏","3 张参考图 + 动作参考视频","一镜到底 · 手机前置视角","变脸梗 · 双人"],"steps":[{"number":1,"title":"第一步：准备 3 张参考图（image_1–3）","description":"image_1、image_2 是两个角色的设定图，每张都要有一张大尺寸脸部特写，全身图上的灰色圆圈是遮挡标记、不能出现在成片里；image_3 是房间图，决定卧室的结构、陈设、配色和光线。作者没有公开这 3 张图，也没有给出图提示词，需要自己准备。"},{"number":2,"title":"第二步：准备动作参考视频（video_1）","description":"前 9 秒的手部动作、表情切换时机、点头和两人站位，全部照搬 video_1。作者在回复里放出了他用的动作视频（一段两人自拍变脸的真人梗视频，16:9 带黑边），提示词里要求成片铺满画面、不要保留参考视频的黑边，也不要继承参考视频里人物的长相和衣服。"},{"number":3,"title":"第三步：先让 GPT 按你的角色改写，再贴进视频模型","description":"作者特别提醒：这段提示词是按他自己的角色和房间写的，不能直接照搬，要先丢给 GPT，让它按你的角色和场景改写。改好后连同 3 张图和动作视频一起交给视频模型。作者没有说明用的是哪个视频模型。"}],"references_detail":[],"storyboard":[{"number":1,"description":"0–0.7s 前置自拍，两人面无表情看镜头；后面的闺蜜一只手举到帽子上方，另一只手放在前面女孩脸下方。"},{"number":2,"description":"0.7–9s 闺蜜两只手上下交替扫过前面女孩的脸，前面的她在咧嘴大笑和闭嘴面无表情之间来回切换，闺蜜始终板着脸。"},{"number":3,"description":"9–10s 闺蜜放下手，两人同时大笑，肩膀抖动，画面轻微晃动，笑着结束。"}],"constraints":"只能有两个人；前面的女孩全程拿着手机，手和手机都不能入镜；一镜到底，不剪辑、不变焦、不绕拍；手扫过时人物的脸、发型、衣服和配饰不能变，帽子不能掉，外套保持露肩；保持参考动作的方向，不要左右镜像；不要镜子画面、多余的手、字幕和换装；最后 1 秒两人的笑声和表情同步。与成片不符：后面闺蜜的手腕上戴着一块表，提示词里没写，像是从动作参考视频带过来的。缺口：作者没公开 image_1–3 这 3 张参考图，也没给出图提示词；成片右上角有一个像素风「G」字水印，看不出是哪个平台；动作参考视频是真人素材，只留在本地，没有放进教程；音频没有核对。","video_prompt":{"title":"双人自拍变脸 · 视频提示词","subtitle":"10s · 1:1 · image_1–3 + video_1 · 英文完整提示词（作者回复长帖）","content":"$31"}},{"id":"nastassiavideo-burger-seedance-2-5","title":"暗调汉堡广告 · 14 镜头快切","subtitle":"X · @nastassiavideo · Seedance 2.5 · 15秒 · 16:9","description":"Seedance 2.5 双层芝士汉堡广告：压肉饼、抛饼、切番茄、淋芝士、一刀插穿，14 个镜头硬切 15 秒。","video":"https://media.liuyidaoai.com/tutorials/nastassiavideo-burger-seedance-2-5/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/nastassiavideo-burger-seedance-2-5/poster.jpg","duration":"15秒","durationSec":15,"styleLabel":"美食广告","shots":14,"references":1,"model":"Seedance 2.5","style":"暗调美食广告 · 微距 + 急推急刹 · 硬切蒙太奇","aspectRatio":"16/9","sourceUrl":"https://x.com/nastassiavideo/status/2100837082243407929","sourceAuthor":"@nastassiavideo","sourcePlatform":"X","sourceImpressions":1240,"sourceStats":{"asOf":"2026-09-27","likes":31,"reposts":1,"bookmarks":4},"formats":["产品广告"],"hook":{"structure":"压肉饼微距 → 人物亮相 → 食材快切 → 组装 → 推板 → 一刀插穿定格","opening":"第 0 秒直接是铁板上铲子把牛肉球压扁的低角度微距，约 2s 切到穿 NastassiaVideo 围裙的她抬眼看镜头。","openingAt":0,"beats":[{"title":"食材快切","text":"约 3–4s 铲起肉饼、手腕一抖把肉饼抛起；约 5s 刀切番茄；约 6s 生菜叶逆光；约 7s 芝士淌过肉饼边缘。","at":3},{"title":"组装成型","text":"约 8s 双层芝士汉堡成品；约 9s 手指把芝麻顶盖按上；约 10–11s 她把木板推向镜头，镜头后退露出人。","at":8},{"title":"结尾怎么收","text":"约 12s 她拿起主厨刀；约 13–14s 刀从顶盖正中插穿整个汉堡；约 15s 刀立在汉堡上，她站在后面看镜头定格。","at":12}],"copyThis":"开头三段先写人物、产品、场景三张参考的文字说明，再逐镜头写「镜头焦段 + 动作 + 镜头怎么加速/刹车 + <音效>」，最后用一段总运镜规则统一节奏。","approx":true},"tags":["15秒 · 美食广告","16:9 横屏","Seedance 2.5","14 镜头逐条写","音效写在尖括号里"],"steps":[{"number":1,"title":"第一步：先用文字写清三张参考","description":"提示词开头三段分别描述人物 Nastassia（灰蓝眼睛、浅棕层次中长发、黑 T 恤、印白字 NastassiaVideo 的黑围裙）、双层芝士汉堡（芝麻布里欧修面包、两块压扁牛肉饼、切达芝士、生菜、番茄、紫洋葱、酸黄瓜、酱，层序固定）和暗调专业厨房，每段末尾挂一个参考图占位符。作者在主帖贴出了人物 Nastassia 的设定图（正脸、侧脸、背面发型、围裙正背面全身），对应第一段的人物参考占位符；汉堡和厨房两张参考图作者没有公开。"},{"number":2,"title":"第二步：定整体影调，再逐镜头写 14 条","description":"总述写 ARRI Alexa 35 质感、24fps、胶片颗粒、90% 画面是深青灰阴影、只留一处琥珀色火光。随后 Shot 1–14 每条都写焦段（24/35/50/85/100mm）、动作、镜头怎么急推急刹，以及尖括号里的音效，比如。"},{"number":3,"title":"第三步：用总运镜规则收尾，贴进 Seedance 2.5","description":"最后一段总规则：冲向动作、在质感上刹车、放进下一刀；广角和微距交替；只在压肉、抛饼最高点和最后一刀用变速；不要环绕、数码变焦、漂浮食材、台词、字幕和片尾字。整段贴进 Seedance 2.5 一次生成。"}],"references_detail":[{"id":"nastassiavideo-burger-nastassia-sheet","number":"1","title":"人物 Nastassia · 设定图","subtitle":"对应提示词第一段人物参考 <<>>；主帖第二张图原件（1600px 压缩）","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-burger-seedance-2-5/refs/nastassia-character-sheet.jpg","prompt":"作者未公开这张设定图的出图提示词。"}],"storyboard":[{"number":1,"description":"0–2s 铁板上铲子压扁牛肉球、铲起煎好的肉饼。"},{"number":2,"description":"2–3s 她站在灶火前抬眼看镜头，微笑。"},{"number":3,"description":"3–5s 铲子铲肉饼、手腕一抖把肉饼抛到空中。"},{"number":4,"description":"5–7s 刀切番茄、逆光生菜叶、芝士从肉饼边缘淌下。"},{"number":5,"description":"8–9s 双层芝士汉堡成品，手指按上芝麻顶盖。"},{"number":6,"description":"10–12s 她把木板推向镜头，拿起主厨刀。"},{"number":7,"description":"13–15s 刀从顶盖插穿汉堡，刀立在汉堡上，她在后面看镜头定格。"}],"constraints":"人物脸和围裙字样、汉堡层序与比例都要和参考一致；变速只用在压肉、抛饼最高点和最后一刀；不要环绕、数码变焦、漂浮食材、镜头穿过实物、台词、人声、字幕、片尾字。与成片不符：提示词写 18 秒、24fps，成片约 15.5 秒、30fps；Shot 6 切紫洋葱在成片里没有出现；Shot 9 两块肉饼落到底座的过程被跳过，约 8s 直接是组装好的汉堡；Shot 2 人物亮相排在第 2 秒，前面只有压肉一镜。缺口：提示词里的汉堡和厨房两张参考图作者没有公开；背景音乐和音效没有逐条核对。","video_prompt":{"title":"Burger Battle · Dark Food Commercial · 14 Shots","subtitle":"Seedance 2.5 · 16:9 · 英文完整提示词（作者贴出的提示词截图，逐字转录）","content":"$32"}},{"id":"nastassiavideo-burj-khalifa-gravity-build-16x9","title":"哈利法塔重力切换建造 · FPV 一镜到底","subtitle":"X · @nastassiavideo · 模型未标注 · 15秒 · 16:9","description":"一镜到底 FPV：哈利法塔从工地地面开始，重力来回切换把钢梁、楼板、玻璃吸上去，15 秒盖完。","video":"https://media.liuyidaoai.com/tutorials/nastassiavideo-burj-khalifa-gravity-build-16x9/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/nastassiavideo-burj-khalifa-gravity-build-16x9/poster.jpg","duration":"15秒","durationSec":15,"styleLabel":"建造过程","shots":1,"references":1,"model":"未标注（作者未公开）","style":"照片级建筑施工特效 · FPV 第一人称飞行 · 一镜到底","aspectRatio":"16/9","sourceUrl":"https://x.com/nastassiavideo/status/2099350859490492465","sourceAuthor":"@nastassiavideo","sourcePlatform":"X","sourceImpressions":14092,"sourceStats":{"asOf":"2026-09-27","likes":99,"reposts":10,"bookmarks":72},"formats":["拆装·制作过程"],"hook":{"structure":"贴地冲刺 → 碎屑上飞 → 钢构猛升 → 立面合拢 → 塔冠失重 → 高空全景","opening":"第 0 秒镜头贴着工地沙地朝远处的楼群冲，约 1s 地面扬起一片粉尘和碎屑。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"约 2–4s 钢梁和碎片往天上飞，镜头跟着钢构往上爬；约 5s 穿过一层还没装立面的混凝土楼板。","at":2},{"title":"立面合拢","text":"约 6–8s 银色立面一圈圈包上塔身，周围全是飞着的板材；约 9–10s 塔冠附近的环形构件在空中漂着一层层落下。","at":6},{"title":"结尾怎么收","text":"约 11–12s 塔尖装好；约 13–15s 镜头拉到高空，整座哈利法塔和青绿色湖水完整入镜。","at":11}],"copyThis":"先写一条「重力方向顺序」（向上 → 向左 → 朝向建筑 → 向上 → 失重 → 向下），再按 0.1 秒精度的时间轴把每次切换对应一段施工阶段，参考图只管最终落成的样子。","approx":true},"tags":["15秒 · 建造过程","16:9 横屏","FPV 一镜到底","中文提示词","1 张建筑参考图"],"steps":[{"number":1,"title":"第一步：准备一张落成后的建筑参考图","description":"作者贴出的参考图是一张竖版高空俯拍的哈利法塔落成图，带迪拜喷泉湖和周边楼群。提示词里用 @image1 引用它，只控制建筑身份和最终完成状态，不限制开场构图。作者在引用帖的回复里说，做法是先用 GPT 2.5 做出完整建筑，再拆成各施工阶段；这张图的出图提示词没有公开。"},{"number":2,"title":"第二步：写规则，再写 0.1 秒精度的时间轴","description":"提示词先规定整体视觉、看不见的 FPV 摄影机、重力切换顺序和「接近 → 旋转 → 对齐 → 接触 → 锁定」的构件规则，然后把 15 秒切成 9 段：贴地突进、猛烈向上、切换向左、朝向建筑、高速装配、再次向上、环绕立面、失重、向下锁定，最后留 2.5 秒完整揭示。"},{"number":3,"title":"第三步：比例选 16:9 生成","description":"提示词原文写的是「竖屏9:16」，作者说同一段提示词出了两版，这版 16:9 是忘了改画幅设置才出来的，她自己更喜欢这版（更有电影感）。跟做时整段照贴，把生成比例设成 16:9 即可。"}],"references_detail":[{"id":"nastassiavideo-burj-image1","number":"1","title":"哈利法塔落成参考图（@image1）","subtitle":"主帖第 2 张图 · 941×1672 竖版高空俯拍","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-burj-khalifa-gravity-build-16x9/refs/image1-burj-khalifa.jpg","prompt":"作者未公开这张参考图的出图提示词（作者在引用帖回复里提到用 GPT 2.5 先做完整建筑）。视频提示词中的用法：@image1 定义最终建筑及其环境的准确外观；参考图控制建筑身份和最终完成状态，不限制开场构图或飞行角度。"}],"storyboard":[{"number":1,"description":"0–2s 镜头贴着工地沙地冲向楼群，地面扬起粉尘碎屑。"},{"number":2,"description":"2–4s 钢梁和碎片向上飞，镜头跟着钢构往上爬，下方露出城市和湖。"},{"number":3,"description":"5s 穿过一层没有立面的混凝土楼板。"},{"number":4,"description":"6–8s 银色立面包上塔身，空中飞满板材。"},{"number":5,"description":"9–12s 塔冠附近的环形构件在空中漂浮、落位，塔尖装好。"},{"number":6,"description":"13–15s 高空斜侧全景，整座哈利法塔和青绿色湖水完整入镜。"}],"constraints":"全程一镜到底，无剪切、无瞬移；画面里不能出现无人机、螺旋桨、摄影机或其影子；镜头只穿过真实的开放缝隙，不穿透实体；重力只作用于松散施工材料，已装结构、城市和湖泊固定；构件不融化、不复制、不凭空生长；无重力波、发光能量、力场图形；只要音效，无对白、音乐、字幕。与成片不符：提示词写「竖屏9:16」，这版成片是 16:9（作者说忘了改画幅）；约 4s 镜头就已经在高空俯看湖面，提示词里这时应该还在工地低处朝向建筑；「切换向左」「失重」这几次重力换向在画面里不容易分辨；成片约 15.1 秒、30fps。缺口：作者没有标注用的视频模型；参考图的出图提示词没公开；提示词截图里有两处字形模糊（「结构跨间」「连续建造进度」），按上下文校读；音频没有核对。","video_prompt":{"title":"哈利法塔——重力切换建造","subtitle":"模型未标注 · 16:9 成片（提示词原文写 9:16）· 中文完整提示词（作者贴出的提示词截图，逐字转录）","content":"$33"}},{"id":"nastassiavideo-grumpy-monster-pumpkin-pie","title":"暴躁怪兽做南瓜派 · 万圣节美食 vlog","subtitle":"X · @nastassiavideo · 模型未标注 · 14秒 · 9:16","description":"毛绒绿怪兽一脸嫌弃地做南瓜派：打蛋、搅拌、压派皮、倒馅、进烤箱，手写字幕美食 vlog，8 镜 15 秒。","video":"https://media.liuyidaoai.com/tutorials/nastassiavideo-grumpy-monster-pumpkin-pie/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/nastassiavideo-grumpy-monster-pumpkin-pie/poster.jpg","duration":"14秒","durationSec":14,"styleLabel":"美食 vlog","shots":8,"references":4,"model":"未标注（作者未公开）","style":"照片级美食博主风 · 万圣节暖调厨房 · 手写动画字幕","aspectRatio":"9/16","sourceUrl":"https://x.com/nastassiavideo/status/2099036524226511264","sourceAuthor":"@nastassiavideo","sourcePlatform":"X","sourceImpressions":1688,"sourceStats":{"asOf":"2026-09-27","likes":41,"reposts":4,"bookmarks":20},"formats":["拆装·制作过程","角色表演"],"hook":{"structure":"怪兽冷脸开场 → 俯拍备料 → 搅拌 → 压派皮 → 倒馅 → 烤箱 → 试吃","opening":"第 0 秒是绿色毛绒怪兽系着南瓜围裙站在空玻璃碗后面，面无表情看镜头，左上角手写「Pumpkin Pie Today!」。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"约 3s 俯拍毛手往碗里打蛋；约 4s 加糖和南瓜；约 5s 搅拌特写配「mix mix mix」；约 6–7s 双手压派皮，配「perfect crust」。","at":3},{"title":"倒馅和烘烤","text":"约 8–9s 玻璃碗把南瓜馅倒进派盘，配「smooth pour」「almost ready」；约 10s 烤箱里的派，配「bake magic」。","at":8},{"title":"结尾怎么收","text":"约 11s 切好的一块派加奶油；约 12–13s 怪兽叉一口尝，脸还是臭着。","at":11}],"copyThis":"用一张四宫格把开场、备料、搅拌、倒馅四个关键画面先做成参考图，再在提示词里逐张写清用途，并锁死同一个碗、同一个派盘。","approx":true},"tags":["14秒 · 美食 vlog","9:16 竖屏","手写动画字幕","4 张参考图","8 镜头按秒分"],"steps":[{"number":1,"title":"第一步：做 4 张关键画面参考图","description":"作者贴出的是一张四宫格：左上怪兽在厨房开场、右上俯拍备料、右下搅拌、左下倒馅，四张都已带手写字幕样式。提示词里分别用 @Image1–@Image4 引用。出图提示词没有公开。"},{"number":2,"title":"第二步：写角色、道具、环境锁定","description":"CHARACTER LOCK 锁怪兽的绿色、稀疏乱毛、耷拉眼、触角和冷脸；PROP LOCK 锁同一个透明玻璃碗和同一个奶油色陶瓷派盘；ENVIRONMENT 锁明亮的万圣节厨房。再规定字幕只能出现在顶部或上侧，用白色手写字加橙色小涂鸦。"},{"number":3,"title":"第三步：按秒写 8 个镜头","description":"0–2s 开场 → 2–3.4s 备料 POV → 3.4–5s 加料 → 5–6.9s 搅拌 → 6.9–8.2s 压派皮 → 8.2–10.2s 低角度倒馅（慢动作只给 0.3–0.4 秒）→ 10.2–11.9s 烤箱 → 11.9–15s 试吃反应，每个镜头写好要出现的字幕。"}],"references_detail":[{"id":"nastassiavideo-pie-image1","number":"1","title":"@Image1 · 怪兽 + 厨房开场","subtitle":"主帖第 2 张四宫格左上 · 已从拼图裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-grumpy-monster-pumpkin-pie/refs/image1-monster-kitchen.jpg","prompt":"作者未公开这张图的出图提示词。视频提示词中的用法：@Image1 = exact monster identity, exact kitchen, exact color palette, exact lighting, exact opening composition."},{"id":"nastassiavideo-pie-image2","number":"2","title":"@Image2 · 俯拍备料 POV","subtitle":"主帖第 2 张四宫格右上 · 已从拼图裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-grumpy-monster-pumpkin-pie/refs/image2-pov-prep.jpg","prompt":"作者未公开这张图的出图提示词。视频提示词中的用法：@Image2 = POV ingredient-prep reference, same glass bowl, same cream ceramic pie dish, same handwritten doodle style."},{"id":"nastassiavideo-pie-image3","number":"3","title":"@Image3 · 搅拌","subtitle":"主帖第 2 张四宫格右下 · 已从拼图裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-grumpy-monster-pumpkin-pie/refs/image3-mixing.jpg","prompt":"作者未公开这张图的出图提示词。视频提示词中的用法：@Image3 = mixing reference, same bowl, same kitchen, same batter texture, same doodle style."},{"id":"nastassiavideo-pie-image4","number":"4","title":"@Image4 · 倒馅","subtitle":"主帖第 2 张四宫格左下 · 已从拼图裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-grumpy-monster-pumpkin-pie/refs/image4-pouring.jpg","prompt":"作者未公开这张图的出图提示词。视频提示词中的用法：@Image4 = pouring reference, same bowl, same cream ceramic pie dish, same kitchen continuity, same doodle style."}],"storyboard":[{"number":1,"description":"0–2s 怪兽站在空玻璃碗后面冷脸看镜头，左上角「Pumpkin Pie Today!」。"},{"number":2,"description":"3–4s 俯拍毛手往碗里打蛋、加糖和南瓜块。"},{"number":3,"description":"5s 搅拌特写，字幕「mix mix mix」。"},{"number":4,"description":"6–7s 双手压派皮，字幕「perfect crust」。"},{"number":5,"description":"8–9s 玻璃碗把南瓜馅倒进派盘，字幕「smooth pour」「almost ready」。"},{"number":6,"description":"10s 烤箱里的南瓜派，字幕「bake magic」。"},{"number":7,"description":"11–14s 一块派加奶油，怪兽叉一口尝，仍然臭脸。"}],"constraints":"怪兽造型不许改；全程同一个玻璃碗、同一个派盘、同一个厨房；字幕不放底部、不挡食物；不要拼贴、分镜格、多余的碗、暗黑恐怖调、过重橙色、畸形手、漂浮食材、随机文字、卡通感。与成片不符：成片约 14.5 秒，提示词写 15 秒；「prep time」「spice it up」「soft & silky」字幕在成片里看不到；结尾没有「spooky good」字幕；试吃时眼神软化不明显。缺口：作者没有标注用的视频模型；4 张参考图的出图提示词没公开；提示词截图里字幕清单有几处字形错乱（如 soft & silky、almost ready、bake magic），按各镜头里的同名字幕校读；音频没有核对。原片 60fps，网页版转成 30fps。","video_prompt":{"title":"Grumpy Monster Makes Pumpkin Pie","subtitle":"模型未标注 · 9:16 · 英文完整提示词（作者贴出的提示词截图，逐字转录）","content":"$34"}},{"id":"nastassiavideo-wrong-prince-kart-race","title":"认错王子 · 卡丁车抢人反转","subtitle":"X · @nastassiavideo · 模型未标注 · 15秒 · 16:9","description":"两个女孩开卡丁车抢王子，冲线后却一起扑向旁边的乌龟怪兽，王子躺地撒泼大哭，15 秒一镜到底。","video":"https://media.liuyidaoai.com/tutorials/nastassiavideo-wrong-prince-kart-race/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/nastassiavideo-wrong-prince-kart-race/poster.jpg","duration":"15秒","durationSec":15,"styleLabel":"喜剧反转","shots":1,"references":5,"model":"未标注（作者未公开）","style":"真人奇幻喜剧 · 卡丁车赛道 · 一镜到底跟拍","aspectRatio":"16/9","sourceUrl":"https://x.com/nastassiavideo/status/2097940769629782288","sourceAuthor":"@nastassiavideo","sourcePlatform":"X","sourceImpressions":2520,"sourceStats":{"asOf":"2026-09-27","likes":52,"reposts":5,"bookmarks":22},"formats":["电影叙事","角色表演"],"hook":{"structure":"赛车抢人 → 飞跃冲线 → 走向王子 → 反转抱怪兽 → 王子撒泼","opening":"第 0 秒是两个女孩开卡丁车的并排脸部特写，戴绿帽和红帽，边开边喊。","openingAt":0,"beats":[{"title":"赛道追逐","text":"约 3–4s 镜头拉远到浮空赛道和金币；约 5s 两辆卡丁车一起飞过断口；约 6–7s 落到城堡前的格子地面刹停。","at":3},{"title":"反转","text":"约 8–11s 王子和乌龟怪兽 Bront 站在前面，两人下车站到他们身边；约 12–13s 两人绕过王子，一左一右抱住 Bront。","at":8},{"title":"结尾怎么收","text":"约 14–15s 王子仰面倒在地上蹬腿撒泼，Bront 搂着两个女孩。","at":14}],"copyThis":"前 8 秒只写赛车、后 7 秒只写喜剧，按 1.5–2.5 秒一段把表情和动作写死（皱鼻子、抿嘴、眼神移到怪兽），反转靠「王子张开双臂、女孩从他身边绕过去」这一个动作完成。","approx":true},"tags":["15秒 · 喜剧反转","16:9 横屏","一镜到底","5 张参考图","马里奥赛车风"],"steps":[{"number":1,"title":"第一步：准备 5 张参考图","description":"作者在帖里放了一段参考图拼图视频：@Image1 浮空赛道世界、@Image2 绿帽女孩 GREEN、@Image3 红帽女孩 RED、@Image4 金色卷发王子、@Image5 乌龟怪兽 Bront。五张都已收录（从拼图视频截帧裁出，约 530px）。出图提示词都没有公开。"},{"number":2,"title":"第二步：写 REFS / GLOBAL / CAMERA","description":"REFS 给每张图写权重（世界 90%，角色 100%）和要保留的特征；GLOBAL 写赛道路线（直道、S 弯、小跳台、断口、下坡、最后一弯、格子线、城堡前广场）和王子与 Bront 等候的位置；CAMERA 规定 84° 广角、一个连续稳定的跟拍运动，从正面脸部特写拉远再侧向弧线移动。"},{"number":3,"title":"第三步：按 6 段时间写动作，贴进生成","description":"0–1.5s 喊「My prince!」「No! MY prince!」→ 1.5–3.5s 超车 → 3.5–5s 飞跃 → 5–8s 漂移冲线刹停 → 8–10s 下车、看到王子皱鼻子、眼神移到 Bront → 10–12.5s 绕开王子亲 Bront → 12.5–15s 王子倒地撒泼大哭。最后写明不要剪切、慢动作、字幕和多余台词。"}],"references_detail":[{"id":"nastassiavideo-prince-image1-world","number":"1","title":"@Image1 · 浮空赛道世界","subtitle":"主帖参考图拼图视频左下格 · 截帧裁出（约 530px）","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-wrong-prince-kart-race/refs/image1-world.jpg","prompt":"作者未公开出图提示词。视频提示词中的用法：@Image1 = world, 90%. Floating roads, mushrooms, pipes, coins, waterfalls, castle."},{"id":"nastassiavideo-prince-image2-green","number":"2","title":"@Image2 · 绿帽女孩 GREEN","subtitle":"主帖参考图拼图视频截帧裁出（约 530px）· AI 角色设定图","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-wrong-prince-kart-race/refs/image2-green.jpg","prompt":"作者未公开出图提示词。视频提示词中的用法：@Image2 = GREEN, 100%. Exact face, dark bob, green cap."},{"id":"nastassiavideo-prince-image3-red","number":"3","title":"@Image3 · 红帽女孩 RED","subtitle":"主帖参考图拼图视频截帧裁出（约 530px）· AI 角色设定图","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-wrong-prince-kart-race/refs/image3-red.jpg","prompt":"作者未公开出图提示词。视频提示词中的用法：@Image3 = RED, 100%. Exact face, light brown hair, red cap."},{"id":"nastassiavideo-prince-image4-prince","number":"4","title":"@Image4 · 金色卷发王子","subtitle":"主帖参考图拼图视频截帧裁出（约 530px）· AI 角色设定图","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-wrong-prince-kart-race/refs/image4-prince.jpg","prompt":"作者未公开出图提示词。视频提示词中的用法：@Image4 = adult prince, 100%. Exact face, blond curls, crown, pink costume."},{"id":"nastassiavideo-prince-image5-bront","number":"5","title":"@Image5 · 乌龟怪兽 Bront","subtitle":"主帖参考图拼图视频左上格 · 截帧裁出（约 530px）","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-wrong-prince-kart-race/refs/image5-bront.jpg","prompt":"作者未公开出图提示词。视频提示词中的用法：@Image5 = Bront, 100%. Exact teal scales, copper shell, horns, white crest, brown eyes."}],"storyboard":[{"number":1,"description":"0–2s 两个女孩开卡丁车的并排脸部特写，边开边喊。"},{"number":2,"description":"3–4s 镜头拉远，浮空赛道、金币和城堡。"},{"number":3,"description":"5s 两辆卡丁车一起飞过断口。"},{"number":4,"description":"6–7s 落到城堡前的格子地面，并排刹停。"},{"number":5,"description":"8–11s 王子和 Bront 站在前面，两个女孩下车站到旁边。"},{"number":6,"description":"12–13s 两人一左一右抱住 Bront。"},{"number":7,"description":"14–15s 王子仰面倒地蹬腿撒泼。"}],"constraints":"角色长相和服装全程一致；参考图的面板排版不能出现在画面里；一镜到底，不要剪切、隐藏剪切、转场、跳时、慢动作和镜头瞬移；不要字幕和提示词以外的台词。与成片不符：约 8s 两人还坐在卡丁车里，王子和 Bront 已在画面中，下车过程不明显；提示词写 GREEN 超车领先，截帧里看不出明确的超车；约 9–11s 两人站在王子和 Bront 两侧，而不是提示词写的面对他们走过来；王子撒泼的蹬腿、捶地只在最后约 1 秒出现；成片约 15.4 秒。缺口：作者没有标注用的视频模型；5 张参考图的出图提示词没公开；参考图只有拼图视频里的低清截帧；台词和音频没有核对。","video_prompt":{"title":"The Wrong Prince","subtitle":"模型未标注 · 16:9 · 英文完整提示词（作者贴出的提示词截图，逐字转录）","content":"$35"}},{"id":"nastassiavideo-pipe-girl-floating-kingdom-parkour","title":"管道女孩 · 浮空王国跑酷","subtitle":"X · @nastassiavideo · Seedance 2.5 · 15秒 · 16:9","description":"红帽背带裤女孩从绿管冲出，踩管子、弹蘑菇、蹬砖墙一路收金币，15 秒一镜到底跑酷。","video":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pipe-girl-floating-kingdom-parkour/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pipe-girl-floating-kingdom-parkour/poster.jpg","duration":"15秒","durationSec":15,"styleLabel":"游戏CG风","shots":1,"references":2,"model":"Seedance 2.5（作者回复标注；同条还写了 Gpt 2、Astra）","style":"游戏过场 CG 风 · 浮空岛跑酷 · 超广角一镜到底后退跟拍","aspectRatio":"16/9","sourceUrl":"https://x.com/nastassiavideo/status/2097311777012805924","sourceAuthor":"@nastassiavideo","sourcePlatform":"X","sourceImpressions":14351,"sourceStats":{"asOf":"2026-09-27","likes":148,"reposts":5,"bookmarks":68},"formats":["角色表演","电影叙事"],"hook":{"structure":"冲出管道 → 管子/蘑菇/砖墙连跳 → 空中收金币 → 翻滚落地 → 胜利姿势","opening":"第 0 秒她从躺倒的绿色大管口里冲出来，身后一圈蓝色闪光，镜头贴着草地往后退。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"约 2s 单手撑竖管翻越；约 3s 跑过横躺的长管；约 4–5s 跳到红蘑菇上方；约 8–10s 在两堵砖墙间蹬墙跑。","at":2},{"title":"空中收金币","text":"约 11–12s 她在空中伸手扫过一串金币，背景是草地浮岛和远处城堡。","at":11},{"title":"结尾怎么收","text":"约 12.3s 落地前滚翻，约 13s 滑铲扬起尘土，约 14–15s 在城堡前举拳定格。","at":12}],"copyThis":"每个动作段都配一句 SFX（金币一声 chime、蘑菇一声闷响），并写一条 coin_rule：碰到的金币当场变成金色火花消失，没碰到的留在空中。","approx":true},"tags":["15秒 · 跑酷","16:9 横屏（另有 9:16 竖版）","一镜到底","Seedance 2.5","马里奥风浮空岛"],"steps":[{"number":1,"title":"第一步：准备两张参考图","description":"@Image1 世界参考（浮空岛、绿管、红蘑菇、砖块、金币、城堡），@Image2 女主角三视图（红帽、奶白背心、牛仔背带裤、白手套、棕靴）。作者回复里写的工具是 Seedance 2.5、Gpt 2、Astra，没说具体哪一步用哪个；出图提示词没公开。两张图都已收录（原图 1672x941，压到 1600px）。"},{"number":2,"title":"第二步：写 REFS / GLOBAL STYLE NOTES / SHOT","description":"REFS 写清每张图只管什么、权重多少（人物 100%、世界和画风 90%），参考图本身不能进画面；GLOBAL 用 medium_lock 锁定「游戏 CG 画风 + 脸接近照片质感」，再写光线、配色比例、场景、角色服装、金币规则、物理和配乐；SHOT 规定 114° 超广角、贴地低机位、全程后退跟拍、第一帧就是满速。"},{"number":3,"title":"第三步：按 12 段写动作和音效","description":"0–1.8s 冲出管道 → 撑竖管翻越 → 跑横管 → 弹第一个蘑菇 → 弹第二个蘑菇 → 最高点收金币 → 蹬砖墙 → 对面砖墙再蹬一次 → 空中收三枚金币 → 翻滚落地 → 滑到平台边 → 13.5–15s 胜利姿势。每段写 dynamic_framing、visual_action、SFX 三行。"},{"number":4,"title":"竖版 9:16（同一条提示词，作者发的第 1 段）","description":"作者同帖还发了一个 720x1280 竖版，用的是同一条游戏 CG 风提示词。竖版原片就没有声音，不是文件坏了。","video":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pipe-girl-floating-kingdom-parkour/demo-vertical-9x16.mp4","poster":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pipe-girl-floating-kingdom-parkour/poster-vertical-9x16.jpg","aspectRatio":"9/16"}],"references_detail":[{"id":"nastassiavideo-pipe-girl-image1-world","number":"1","title":"@Image1 · 浮空岛世界主视觉","subtitle":"作者回复区图片原件 HRtzI--WcAArnjm（1600px 压缩）· 画面里女主从绿管口飞出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pipe-girl-floating-kingdom-parkour/refs/image1-world-keyvisual.jpg","prompt":"作者未公开出图提示词。视频提示词中的用法：@Image1 = world reference. Controls the floating-island geography, green pipes, red mushroom caps, brick blocks, coin trails, waterfalls, castle; render style 90%."},{"id":"nastassiavideo-pipe-girl-image2-woman","number":"2","title":"@Image2 · 女主三视图","subtitle":"作者回复区图片原件 HRtzI-6XYAAHlom（1600px 压缩）· AI 角色设定图","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pipe-girl-floating-kingdom-parkour/refs/image2-woman-charsheet.jpg","prompt":"作者未公开出图提示词。视频提示词中的用法：@Image2 = the Woman. Controls face, body proportions and wardrobe only (identity 100%)."}],"storyboard":[{"number":1,"description":"0–1s 从绿管口冲出，拖着蓝色闪光冲向镜头。"},{"number":2,"description":"2s 单手撑竖管翻越。"},{"number":3,"description":"3s 在横躺的长绿管上奔跑。"},{"number":4,"description":"4–5s 腾空，跳向红蘑菇。"},{"number":5,"description":"5.5–7.5s 从闪光管口前飞过，两侧是红蘑菇，远处城堡。"},{"number":6,"description":"8–10s 在两堵砖墙之间蹬墙跑，砖屑飞溅。"},{"number":7,"description":"11–12s 空中伸手收一串金币。"},{"number":8,"description":"12.3–13s 前滚翻落地，滑铲扬尘。"},{"number":9,"description":"14–15s 城堡前举拳定格。"}],"constraints":"人物长相、发色、服装全程一致；参考图的姿势和三视图面板不能进画面；一镜到底，镜头始终在她前方后退。碰到的金币必须消失，没碰到的留在空中。与成片不符：约 5.5–7.5s 她几乎定格在「从闪光管口飞出、两侧红蘑菇」的同一个姿势，和作者贴出的第一张横版图构图几乎一样，提示词写的第二次弹蘑菇、蘑菇压扁回弹看不清；两次蹬墙之间的过渡不明显；成片约 15.1 秒。缺口：作者没说 Seedance 2.5、Gpt 2、Astra 各用在哪一步；参考图出图提示词没公开；音频只量了音量，没有逐段核对音效。","video_prompt":{"title":"Floating Kingdom Pipe Run（游戏 CG 风版）","subtitle":"Seedance 2.5 · 16:9 主成片 + 9:16 竖版共用 · 英文完整提示词（作者回复里的提示词截图，逐字转录）","content":"$36"}},{"id":"nastassiavideo-medianoche-saloon-escape","title":"蒙面女侠 · 酒馆打斗跳马追马","subtitle":"X · @nastassiavideo · Seedance 2.5 · 32秒 · 16:9","description":"蒙面女侠在酒馆轻松打趴三个劫匪，撞窗逃出跳上黑马，却被树枝扫落，只能追着马跑。","video":"https://media.liuyidaoai.com/tutorials/nastassiavideo-medianoche-saloon-escape/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/nastassiavideo-medianoche-saloon-escape/poster.jpg","duration":"32秒","durationSec":32,"styleLabel":"复古喜剧","shots":11,"references":1,"model":"Seedance 2.5（作者帖文：Seedance 2.5、GPT-2、Claude）","style":"1930–40 年代特艺色冒险片 · 西部酒馆 · 肢体喜剧","aspectRatio":"16/9","sourceUrl":"https://x.com/nastassiavideo/status/2095368017739653337","sourceAuthor":"@nastassiavideo","sourcePlatform":"X","sourceImpressions":4162,"sourceStats":{"asOf":"2026-09-27","likes":33,"reposts":3,"bookmarks":9},"formats":["电影叙事","角色表演"],"hook":{"structure":"酒馆打斗 → 撞见追兵 → 撞窗逃跑 → 吹哨跳马 → 被树枝扫落追马","opening":"第 0 秒劫匪的剑从画面左侧刺过来，蒙面女侠在昏暗的酒馆里单手格挡。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"约 2.5–4s 撑桌翻越，劫匪扑空滑过桌面；约 4–6.5s 一脚把第三个劫匪踢进木桶堆；约 9–11.5s 她的脸部近景，身后门口进来一群持枪的人。","at":2.5},{"title":"逃跑上马","text":"约 14.5s 肩膀撞碎百叶窗冲到街上；约 16s 两指放进嘴里吹口哨；约 17–19.5s 黑马跑来刹停，她跳上马背。","at":14.5},{"title":"结尾怎么收","text":"约 23–25.5s 迎面的树枝把她从马上扫下来；约 26–29s 她爬起来追马；约 30–32s 远景圆形遮罩收成黑场。","at":23}],"copyThis":"每个镜头都写到 0.1 秒的时间码，并在动作之后塞一个短表情反应（挑眉歪嘴、瞪眼咽口水、咬紧牙），再用 HARD CUT / WHIP PAN / SMASH CUT 连起来，喜剧节奏靠这些反应停顿。","approx":true},"tags":["32秒 · 西部喜剧","16:9 横屏","11 个镜头","Seedance 2.5","复古胶片圆形遮罩"],"steps":[{"number":1,"title":"第一步：准备角色设定图","description":"作者在回复里贴了角色 Medianoche 的设定图（正脸、侧脸、背面、全身正背面，外加身高 158cm、声音、性格说明），就是提示词里的 @Image1。帖文写的工具是 Seedance 2.5、GPT-2、Claude，没说各用在哪一步。设定图已收录（原图 1672x941，压到 1600px）。"},{"number":2,"title":"第二步：先写各种「锁」","description":"REFS 锁脸和服装；FORMAT 定死 11 个镜头、30 秒；再写 GENRE AND COMEDY、PACING LOCK（每个镜头 0.2 秒内开始动作、打斗像 18fps 拍 24fps 放）、MIMICRY LOCK（挑眉、歪嘴笑、瞪眼等 8 种微表情）、VISUAL STYLE、场景 / 道具 / 马的连贯性。"},{"number":3,"title":"第三步：11 个镜头逐个写时间码和转场","description":"每个镜头写景别、机位、精确到 0.1 秒的动作、SFX，以及 HARD CUT / WHIP PAN / AUDIO BRIDGE / SMASH CUT 等转场词。跳马那一镜作者单独写了方向锁（要求面朝马尾反坐）；结尾写明 29.5 秒开始圆形遮罩、30.0 秒刚好黑场。最后用 FINAL CONSTRAINTS 把所有要求再列一遍。"}],"references_detail":[{"id":"nastassiavideo-medianoche-escape-image1","number":"1","title":"@Image1 · Medianoche 角色设定图","subtitle":"作者回复区图片原件 HRQ-HUvXsAM3f4O（1600px 压缩）· AI 角色设定图","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-medianoche-saloon-escape/refs/image1-medianoche-charsheet.jpg","prompt":"作者未公开出图提示词。视频提示词中的用法：@Image1 is the only character reference for Medianoche — same face, facial geometry, body proportions, hat, mask and cape in every shot."}],"storyboard":[{"number":1,"description":"0–2.5s 酒馆里单手格挡劫匪的剑，收剑挑眉。"},{"number":2,"description":"2.5–4s 撑桌翻越，劫匪扑空滑过桌面。"},{"number":3,"description":"4–7s 踢飞第三个劫匪撞进木桶堆。"},{"number":4,"description":"7.5–9s 酒馆中景，她站在桌边。"},{"number":5,"description":"9–11.5s 脸部近景，身后门口涌进一群人。"},{"number":6,"description":"12–14s 在桌椅间躲开追兵往窗口跑。"},{"number":7,"description":"14.5–15.5s 撞碎百叶窗冲到街上。"},{"number":8,"description":"16–16.5s 街上近景，两指放进嘴里吹口哨。"},{"number":9,"description":"17–19.5s 黑马跑来刹停，她跳上马背。"},{"number":10,"description":"20–22.5s 从马后方拍她骑马跑远；23s 转为侧面跟拍。"},{"number":11,"description":"23–26s 迎面的树枝把她扫落，摔在土路上。"},{"number":12,"description":"26.5–29s 背影追马。"},{"number":13,"description":"29.5–32s 高角度远景，圆形遮罩收成黑场。"}],"constraints":"Medianoche 的脸、面具、帽子、披风全程不变；披风不能掉，第一镜后剑一直插在右腰剑鞘里；马跑来时背上没人，完全停下后她才跳上去；她必须面朝马尾，树枝打在后背和肩膀；不要字幕、Logo 和提示词以外的台词；喊「Wait! Wait for me!」时不要背景音乐。与成片不符：提示词最核心的笑点「面朝马尾反坐」在成片里没有做出来，约 23–24.5s 的侧面镜头里她面朝马头方向坐着，树枝是从正面扫到她身上，不是打在后背；提示词写的踩到木板滑一跤看不清；成片约 32 秒，比提示词写的 30 秒长，各镜头整体后移约 1.5–2 秒；圆形遮罩约 30–32 秒才收完。缺口：作者没说 Seedance 2.5、GPT-2、Claude 各用在哪一步；角色设定图的出图提示词没公开；提示词原图只有 339×2048，破折号、引号的具体字形看不清。","video_prompt":{"title":"Medianoche · Playful Fight to Backward Mount","subtitle":"Seedance 2.5 · 16:9 · 英文完整提示词（作者回复里的提示词长图，逐字转录）","content":"$37"}},{"id":"nastassiavideo-medianoche-saloon-porch-gag","title":"蒙面女侠 · 酒馆门廊出糗","subtitle":"X · @nastassiavideo · Seedance 2.5 · 19秒 · 16:9","description":"蒙面女侠帅气推门出场却扑倒在地，起身装没事，躲拳让劫匪互殴，最后捡帽退场。","video":"https://media.liuyidaoai.com/tutorials/nastassiavideo-medianoche-saloon-porch-gag/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/nastassiavideo-medianoche-saloon-porch-gag/poster.jpg","duration":"19秒","durationSec":19,"styleLabel":"复古喜剧","shots":6,"references":1,"model":"Seedance 2.5（帖文：视频 Seedance 2.5，角色卡 GPT 2）","style":"早期彩色冒险片 · 西部酒馆门廊 · 肢体喜剧","aspectRatio":"16/9","sourceUrl":"https://x.com/nastassiavideo/status/2096952053687808223","sourceAuthor":"@nastassiavideo","sourcePlatform":"X","sourceImpressions":2591,"sourceStats":{"asOf":"2026-09-27","likes":31,"reposts":4,"bookmarks":13},"formats":["角色表演","电影叙事"],"hook":{"structure":"帅气出场 → 绊倒 → 装没事 → 躲拳误伤 → 披风绊倒一堆 → 捡帽退场","opening":"第 0–1.5 秒酒馆双开门被推开，蒙面女侠披着斗篷大步走出来；约 2 秒她扑倒向镜头，帽子飞到镜头前。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"约 3–5s 近景里她扶正帽子和眼罩，抬下巴装作一切顺利；约 5.5–7s 劫匪挥拳，她低头躲过，拳头打在后面另一个劫匪身上。","at":3},{"title":"披风绊倒一堆","text":"约 8.5–12s 高角度门廊全景：她拔剑，披风甩开，披条纹披毯的劫匪扑过来，最后几个人摔在木桶旁。","at":8.5},{"title":"结尾怎么收","text":"约 13–15.5s 脸部特写挑眉；约 16–17.5s 收剑从画面左边走出；约 18–19s 一只黑手套拿着帽子伸进画面，圆形遮罩收成黑场。","at":13}],"copyThis":"每个镜头都用 vfx / camera_motion / action_visual / exit 四行写，exit 写清这一镜最后一个动作和转场方式（HARD CUT、MATCH CUT、WHIP PAN），镜头之间就能接得顺。","approx":true},"tags":["19秒 · 西部喜剧","16:9 横屏","6 个镜头","Seedance 2.5","复古胶片圆形遮罩"],"steps":[{"number":1,"title":"第一步：准备角色卡","description":"@Image1 用的是同一张 Medianoche 角色卡（正脸、侧脸、背面、全身正背面，身高 158cm、声音、性格说明），帖文写角色卡用 GPT 2 做。角色卡已收录（原图 1672x941，压到 1600px），出图提示词没公开。"},{"number":2,"title":"第二步：写全局设定和 4 个角色","description":"先写一句 @Image1 要保留的所有外观细节，再写 SCENE 的情节线（出场 → 出糗 → 打斗 → 淡定 → 退场笑点）、SHOT STRUCTURE（6 个镜头、20 秒）、GLOBAL STYLE NOTES（正午硬光、早期彩色片调色、门廊布景、环境音），最后给女主和 3 个劫匪各写一行 visual_anchors。"},{"number":3,"title":"第三步：6 个镜头按四行格式写","description":"HOOK 出场摔倒 → PUNCHLINE 起身装没事 → RISING 躲拳误伤 → SIGNATURE 剑勾披风绊倒三人 → CALLBACK 特写挑眉 → RESOLUTION 收剑、退场、捡帽、圆形遮罩。结尾用 GENERATE FIRST 写明先生成第 4、5、6 镜。"}],"references_detail":[{"id":"nastassiavideo-medianoche-porch-image1","number":"1","title":"@Image1 · Medianoche 角色卡","subtitle":"作者帖内图片原件 HRnetSpbkAEnwzc（1600px 压缩）· 与《酒馆脱逃》同一张 AI 角色卡","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-medianoche-saloon-porch-gag/refs/image1-medianoche-charsheet.jpg","prompt":"作者未公开出图提示词（帖文写角色卡用 GPT 2 做）。视频提示词中的用法：@Image1 is the masked heroine reference — hat, eye mask, cape, corset bodice, trousers, gloves, boots."}],"storyboard":[{"number":1,"description":"0–1.5s 酒馆双开门推开，女侠大步走出。"},{"number":2,"description":"2s 扑倒向镜头，帽子飞到镜头前。"},{"number":3,"description":"3–5s 近景，扶正帽子和眼罩。"},{"number":4,"description":"5.5–8s 劫匪挥拳，她低头躲开，拳头打中另一个劫匪。"},{"number":5,"description":"8.5–12s 高角度门廊全景，拔剑甩披风，劫匪们摔在木桶旁。"},{"number":6,"description":"13–15.5s 脸部特写，身后劫匪虚焦。"},{"number":7,"description":"15.5–17.5s 收剑，从画面左边走出。"},{"number":8,"description":"18–19s 黑手套拿着帽子伸进画面，圆形遮罩收成黑场。"}],"constraints":"女主的帽子、眼罩、披风、紧身胸衣、长靴、右腰的剑每个镜头都一样；3 个劫匪按各自服装区分（小胡子背带、棕马甲红领巾、条纹披毯）；全程复古胶片质感。与成片不符：提示词写 4:3、20 秒，成片是 16:9、约 19.3 秒；剑勾住披风、披风盖住劫匪头的过程在高角度全景里看不清；提示词写捡回帽子戴到头上，成片最后只看到手拿着帽子伸进画面；收剑两次插不进没有明显表现。缺口：角色卡出图提示词没公开。","video_prompt":{"title":"Medianoche · Heroic Entrance to Exit Gag","subtitle":"Seedance 2.5 · 提示词写 4:3（成片 16:9）· 英文完整提示词（主帖第二张图，逐字转录）","content":"$38"}},{"id":"nastassiavideo-suede-bag-ugc-review","title":"麂皮手提包 · UGC 开箱口播","subtitle":"X · @nastassiavideo · 模型未标注 · 20秒 · 9:16","description":"博主坐在床边拆盒取出麂皮手提包，边讲边展示缝线和内袋，再到镜子前试背，20 秒种草口播。","video":"https://media.liuyidaoai.com/tutorials/nastassiavideo-suede-bag-ugc-review/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/nastassiavideo-suede-bag-ugc-review/poster.jpg","duration":"20秒","durationSec":20,"styleLabel":"真人风","shots":1,"references":4,"model":"未标注（作者未公开）","style":"UGC 博主测评 · 卧室暖光 · 手机手持竖拍","aspectRatio":"9/16","sourceUrl":"https://x.com/nastassiavideo/status/2096179182992527601","sourceAuthor":"@nastassiavideo","sourcePlatform":"X","sourceImpressions":1990,"sourceStats":{"asOf":"2026-09-27","likes":50,"reposts":5,"bookmarks":22},"formats":["产品广告","手机POV·Vlog"],"hook":{"structure":"拆盒 → 细节展示 → 打开看内袋 → 镜前试背 → 举包收尾","opening":"第 0 秒她坐在床边，腿上放着一个白色纸盒，对着镜头笑着开口说话。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"约 1–4s 打开盒子取出包；约 4.5–8.5s 把包横在镜头前转动，展示前片、明线和手柄；约 9–11.5s 打开磁扣，把包口朝向镜头看内袋。","at":1},{"title":"上身效果","text":"约 12–15.5s 她站起来把包背上肩，走到落地镜前侧身照。","at":12},{"title":"结尾怎么收","text":"约 16–17.5s 回到床边把包放下再拿起；约 18–20s 把包举到镜头前说最后一句。","at":16}],"copyThis":"把商品参数写死：材质（仿麂皮、哑光绒面）、配件（两根细圆手柄、两个前片小搭扣、磁扣、内侧拉链袋）和尺寸（37×17.5×10.5 cm），再让每句台词对应一个展示动作。","approx":true},"tags":["20秒 · 带货口播","9:16 竖屏","一镜到底","商品参考图","UGC 开箱"],"steps":[{"number":1,"title":"第一步：准备人物图和商品图","description":"作者在主帖放了一张 2×2 拼图：三张包的棚拍图（斜侧、内部、正面）和一张女性面部特写。四张都已收录（面部特写是 AI 生成的人物参考）。作者没有标注用的视频模型。"},{"number":2,"title":"第二步：锁定人物和商品","description":"第一段写人物参考要保留的脸、发型、妆容和服装（奶白针织上衣、直筒牛仔裤、细金饰）；第二段把包的材质、颜色、手柄、搭扣、磁扣、内袋和尺寸全部写死，要求每个镜头都一样。"},{"number":3,"title":"第三步：按台词顺序写动作","description":"场景写明卧室、黄金时段暖光、手机手持轻微晃动。之后按顺序写 5 句台词，每句配一个动作：拆盒、转包看细节、打开磁扣看内袋、镜前试背、看镜头说收尾句。最后写声音（温暖的中音）、画面质感和「不要字幕、Logo、水印」。"}],"references_detail":[{"id":"nastassiavideo-suede-bag-image1-angle","number":"1","title":"手提包 · 斜侧面","subtitle":"主帖拼图左上格裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-suede-bag-ugc-review/refs/image1-bag-angle.jpg","prompt":"作者上传的商品图，不是生成图，没有出图提示词。"},{"id":"nastassiavideo-suede-bag-image2-interior","number":"2","title":"手提包 · 内部与内袋","subtitle":"主帖拼图左下格裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-suede-bag-ugc-review/refs/image2-bag-interior.jpg","prompt":"作者上传的商品图，不是生成图，没有出图提示词。"},{"id":"nastassiavideo-suede-bag-image3-front","number":"3","title":"手提包 · 正面","subtitle":"主帖拼图右下格裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-suede-bag-ugc-review/refs/image3-bag-front.jpg","prompt":"作者上传的商品图，不是生成图，没有出图提示词。"},{"id":"nastassiavideo-suede-bag-image4-face","number":"4","title":"人物参考 · 面部特写","subtitle":"主帖拼图右上格裁出（512x540）· AI 人物参考","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-suede-bag-ugc-review/refs/image4-face-closeup.jpg","prompt":"作者未公开出图提示词。提示词第一段按这张图锁定脸、发型、妆容。"}],"storyboard":[{"number":1,"description":"0–1s 坐在床边，腿上放白盒子，看镜头开口。"},{"number":2,"description":"1–4s 打开盒子，取出麂皮包。"},{"number":3,"description":"4.5–8.5s 把包横在镜头前转动，展示前片和明线。"},{"number":4,"description":"9–11.5s 打开磁扣，展示内部和拉链内袋。"},{"number":5,"description":"12–13s 站起来把包背上肩。"},{"number":6,"description":"13.5–15.5s 在落地镜前侧身照。"},{"number":7,"description":"16–17.5s 回到床边，把包放到盒子旁再拿起。"},{"number":8,"description":"18–20s 把包举到镜头前说收尾句。"}],"constraints":"人物脸、发型、妆容、服装全程一致；包的形状、颜色、绒面、明线、手柄、磁扣每个镜头都一样；不要字幕、Logo、水印。与成片不符：提示词写不要字幕，成片全程有逐词弹出的大字幕，应该是后期加的；结尾是她把包举到镜头前，不是提示词写的镜头推近后淡出。缺口：作者没有标注视频模型；口型和台词没有逐句核对。","video_prompt":{"title":"Suede Shoulder Bag · UGC Review","subtitle":"模型未标注 · 9:16 · 英文完整提示词（主帖第一张图，逐字转录）","content":"$39"}},{"id":"nastassiavideo-pink-after-dark-rap-minimax-h3","title":"粉色夜城说唱 MV · MiniMax H3","subtitle":"X · @nastassiavideo · MiniMax H3 · 12秒 · 16:9","description":"粉色霓虹夜城说唱 MV：每句歌词换一个雨夜场景，大字歌词印在她身后的墙、围栏和楼上。","video":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pink-after-dark-rap-minimax-h3/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pink-after-dark-rap-minimax-h3/poster.jpg","duration":"12秒","durationSec":12,"styleLabel":"真人风","shots":11,"references":10,"model":"MiniMax H3","style":"雨夜霓虹 · 酸性玫瑰粉 + 黑 · 35mm 胶片颗粒 · 巨型歌词排版","aspectRatio":"16/9","sourceUrl":"https://x.com/nastassiavideo/status/2093364039187308582","sourceAuthor":"@nastassiavideo","sourcePlatform":"X","sourceImpressions":2020,"sourceStats":{"asOf":"2026-09-27","likes":33,"reposts":4,"bookmarks":10},"formats":["字效·片头","角色表演"],"hook":{"structure":"每句歌词一个场景 → 歌词大字落在她身后的真实表面 → 卡在拍点上硬切","opening":"第 0 秒她站在砖墙小巷里，身后一道竖向粉色霓虹；0.25 秒起「PINK AFTER DARK」三行大字从积水倒影里升起，贴到左边墙上。","openingAt":0,"beats":[{"title":"场景怎么换","text":"约 1.75s 切到地铁站近景，她整理马甲领口；2.5s 砖墙上贴出「CITY ON MY SKIN」海报；3.25s 铁丝网横幅「YOU CAN'T PULL ME IN」；3.75s 雨夜斑马线「I WALK LIKE THUNDER」，脚下一圈粉色涟漪。","at":1.75},{"title":"中段","text":"约 5s 天台粉烟里浮出「BLACK SKY BURNING」；6.25s 地下车库，「HEARTBEAT IN THE STREET」外套银色心电圆环；7.25s 玻璃大楼窗格拼出「I OWN THE NIGHT」，她摸头顶墨镜，随后窗格一排排熄灭。","at":5},{"title":"结尾怎么收","text":"约 8.5s 地下通道壁画上斜着落下「MOVE TO MY RHYTHM」；9.75s 切到隧道口；10–12.3s 近景，身后巨型「CONCRETE」和小字「WATCH ME LIGHT THE」，她看着镜头唱完最后一句。","at":8.5}],"copyThis":"把字写成场景的一部分：规定字体（超窄粗黑体、粉色填充 + 细银边）、字在她身后的哪种表面上（墙、横幅、窗格、壁画），并要求字离镜头比人远、不挡脸、拼写正确。","approx":true},"tags":["12秒 · 说唱 MV","16:9 横屏","11 个镜头","歌词大字排版","角色卡 + 音频驱动"],"steps":[{"number":1,"title":"第一步：准备角色卡、歌词场景图和音轨","description":"作者在主帖放了三张图：一张角色卡（全身 + 脸部特写 + 半身，粉色马甲造型）、一张 3×3 歌词场景图（9 个场景各配一句歌词大字），和一张提示词长图。角色卡和 9 张场景图都已收录。提示词里的 @Audio1 是说唱音轨，作者没有公开。"},{"number":2,"title":"第二步：写一段三次生成都用的公共设定","description":"开头的 SHARED BLOCK 每次生成都贴：@Image1 只管脸、发色、妆和服装，@Audio1 只管节奏和口型；每个场景只有一个酸性玫瑰粉霓虹光源、积水从下往上补光；色调约 70% 黑 + 27% 粉 + 3% 银；写明服装细节、字体规则和机位范围；音乐 160 BPM、每个切点落在小节第一拍、不要模型自己生成音乐。"},{"number":3,"title":"第三步：分三次生成，每次 3 个镜头","description":"RUN A、B、C 各写 3 个镜头（5 秒、5 秒、6 秒），每个镜头都写：机位和镜头焦段、灯光变化、运镜、动作和要唱的歌词、字怎么出现、怎么转到下一个镜头（硬切、甩镜、光扫、烟雾转场、匹配剪辑、推镜爆闪）。结尾统一贴技术限制：只出现列出的字、不要字幕、水印、Logo、多余的人、不要换装换脸。"}],"references_detail":[{"id":"nastassiavideo-pink-after-dark-image1-nastya","number":"@Image1","title":"@Image1 · 角色卡 Nastya","subtitle":"主帖第一张图原件 HQ0fbK0WUAAsL00（1536x1024）· 全身 + 脸部特写 + 半身，AI 角色卡","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pink-after-dark-rap-minimax-h3/refs/image1-character-card-nastya.jpg","prompt":"作者未公开出图提示词。视频提示词中的用法：@Image1 = character card. Controls face, grey blue eyes, ash brown hair, makeup and wardrobe only; the backdrop of @Image1 stays out of the output."},{"id":"nastassiavideo-pink-after-dark-map1","number":"1","title":"歌词场景图 · PINK AFTER DARK","subtitle":"砖墙小巷 + 竖向粉色霓虹；主帖第二张 3×3 拼图裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pink-after-dark-rap-minimax-h3/refs/map1-pink-after-dark.jpg","prompt":"作者没有公开这张图的出图提示词。"},{"id":"nastassiavideo-pink-after-dark-map2","number":"2","title":"歌词场景图 · CITY ON MY SKIN","subtitle":"黑砖墙 + 三条粉色海报；主帖第二张 3×3 拼图裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pink-after-dark-rap-minimax-h3/refs/map2-city-on-my-skin.jpg","prompt":"作者没有公开这张图的出图提示词。"},{"id":"nastassiavideo-pink-after-dark-map3","number":"3","title":"歌词场景图 · I WALK LIKE THUNDER","subtitle":"雨夜斑马线 + 粉色涟漪；主帖第二张 3×3 拼图裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pink-after-dark-rap-minimax-h3/refs/map3-i-walk-like-thunder.jpg","prompt":"作者没有公开这张图的出图提示词。"},{"id":"nastassiavideo-pink-after-dark-map4","number":"4","title":"歌词场景图 · YOU CAN'T PULL ME IN","subtitle":"铁丝网 + 粉色横幅；主帖第二张 3×3 拼图裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pink-after-dark-rap-minimax-h3/refs/map4-you-cant-pull-me-in.jpg","prompt":"作者没有公开这张图的出图提示词。"},{"id":"nastassiavideo-pink-after-dark-map5","number":"5","title":"歌词场景图 · BLACK SKY BURNING","subtitle":"天台 + 粉色烟雾；主帖第二张 3×3 拼图裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pink-after-dark-rap-minimax-h3/refs/map5-black-sky-burning.jpg","prompt":"作者没有公开这张图的出图提示词。"},{"id":"nastassiavideo-pink-after-dark-map6","number":"6","title":"歌词场景图 · HEARTBEAT IN THE STREET","subtitle":"地下车库 + 银色圆环心电线；主帖第二张 3×3 拼图裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pink-after-dark-rap-minimax-h3/refs/map6-heartbeat-in-the-street.jpg","prompt":"作者没有公开这张图的出图提示词。"},{"id":"nastassiavideo-pink-after-dark-map7","number":"7","title":"歌词场景图 · I OWN THE NIGHT","subtitle":"玻璃大楼窗格亮字；主帖第二张 3×3 拼图裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pink-after-dark-rap-minimax-h3/refs/map7-i-own-the-night.jpg","prompt":"作者没有公开这张图的出图提示词。"},{"id":"nastassiavideo-pink-after-dark-map8","number":"8","title":"歌词场景图 · MOVE TO MY RHYTHM","subtitle":"地下通道几何壁画；主帖第二张 3×3 拼图裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pink-after-dark-rap-minimax-h3/refs/map8-move-to-my-rhythm.jpg","prompt":"作者没有公开这张图的出图提示词。"},{"id":"nastassiavideo-pink-after-dark-map9","number":"9","title":"歌词场景图 · WATCH ME LIGHT THE CONCRETE","subtitle":"隧道口粉色天际 + 巨型字；主帖第二张 3×3 拼图裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-pink-after-dark-rap-minimax-h3/refs/map9-watch-me-light-the-concrete.jpg","prompt":"作者没有公开这张图的出图提示词。"}],"storyboard":[{"number":1,"description":"0–1.6s 砖墙小巷，竖向粉色霓虹；「PINK AFTER DARK」从积水倒影里升起贴到墙上。"},{"number":2,"description":"1.75–2.4s 地铁站近景，背后列车掠过，她整理马甲领口。"},{"number":3,"description":"2.5–3.1s 黑砖墙，「CITY / ON MY / SKIN」三条粉色海报贴在她身后。"},{"number":4,"description":"3.25–3.6s 铁丝网前，粉色横幅「YOU CAN'T PULL ME IN」。"},{"number":5,"description":"3.75–4.9s 雨夜斑马线，她朝镜头走来，「I WALK LIKE THUNDER」斜体大字，脚下粉色涟漪扩散。"},{"number":6,"description":"5–5.9s 天台低角度，粉色烟雾里浮出「BLACK SKY BURNING」，风吹起头发。"},{"number":7,"description":"6–7.1s 地下车库粉色灯管，「HEARTBEAT IN THE STREET」外套银色圆环和心电线。"},{"number":8,"description":"7.25–8.3s 玻璃大楼窗格拼出「I OWN THE NIGHT」，她摸头顶墨镜，窗格一排排熄灭。"},{"number":9,"description":"8.5–9.6s 地下通道几何壁画，「MOVE / TO MY / RHYTHM」斜着落下，她甩手腕转身。"},{"number":10,"description":"9.75s 隧道口，远处粉色天际，隧道灯依次亮起。"},{"number":11,"description":"10–12.3s 近景，身后巨型「CONCRETE」和小字「WATCH ME LIGHT THE」，她看镜头唱完最后一句。"}],"constraints":"只出现列出的字，拼写正确，不要字幕、水印、Logo；字不挡脸；只有她一个人，不换装不换脸；不要模型自己生成音乐。与成片不符：提示词写 9:16、分三次生成（5+5+6 秒），成片是一条 12 秒 16:9 横屏；成片里的字是整句歌词（如「YOU CAN'T PULL ME IN」），提示词只写了部分关键词（如「PULL ME IN」）；「CITY ON MY SKIN」在成片里是地铁近景后另切一个砖墙镜头，不是贴在地铁瓷砖上；成片把铁丝网镜头放在斑马线之前。缺口：@Audio1 音轨没有公开；9 张歌词场景图怎么用（是否作为参考图上传）作者没说；口型和歌词没有逐句核对。","video_prompt":{"title":"Pink After Dark · Rap Music Video","subtitle":"MiniMax H3 · 提示词写 9:16（成片 16:9）· 英文完整提示词（主帖第三张图，逐字转录）","content":"$3a"}},{"id":"nastassiavideo-brown-loafers-unboxing-seedance-2-5","title":"棕色乐福鞋 · 开箱试穿种草 · Seedance 2.5","subtitle":"X · @nastassiavideo · Seedance 2.5 · 23秒 · 9:16","description":"棕色乐福鞋开箱种草：拆盒、看细节、试穿、走到镜头前，手机手持拍法，配口播和大字幕。","video":"https://media.liuyidaoai.com/tutorials/nastassiavideo-brown-loafers-unboxing-seedance-2-5/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/nastassiavideo-brown-loafers-unboxing-seedance-2-5/poster.jpg","duration":"23秒","durationSec":23,"styleLabel":"真人风","shots":10,"references":5,"model":"Seedance 2.5","style":"卧室晨光 · 暖中性色 + 棕色皮革 + 靛蓝牛仔 · 手机手持 UGC","aspectRatio":"9/16","sourceUrl":"https://x.com/nastassiavideo/status/2092993847475257425","sourceAuthor":"@nastassiavideo","sourcePlatform":"X","sourceImpressions":1826,"sourceStats":{"asOf":"2026-09-27","likes":33,"reposts":2,"bookmarks":7},"formats":["产品广告","手机POV·Vlog"],"hook":{"structure":"抱盒开口 → 拆盒 → 手持看鞋 → 微距看细节 → 穿上 → 俯视看脚 → 走到镜头前","opening":"第 0 秒她坐在床边，把米色鞋盒捧到胸前，对着镜头说「Okay, these finally came」。","openingAt":0,"beats":[{"title":"拆盒和看鞋","text":"约 3–6.3s 俯拍掀盖、再切到她坐在地板上翻开白色包装纸；6.5–8.3s 一只手把鞋举起，四分之三侧面对镜头。","at":3},{"title":"细节","text":"约 8.5–10.8s 微距，指尖沿着鞋面横带和冲孔滑过；11–12.3s 翻过来看黑色锯齿大底；12.5–14.8s 双手拿鞋举到镜头前。","at":8.5},{"title":"上脚和收尾","text":"约 15–17.3s 低机位看脚伸进鞋里踩实；17.5–19.8s 站着往下拍两只鞋；20–23.4s 低机位全身，她从左边走进来站定面对镜头。","at":15}],"copyThis":"把鞋的四个视角分别当参考图，并写清每张图对应哪个角度；每个镜头都写机位、焦段、手在哪、光从哪来，让手持手机的质感贯穿全片。","approx":true},"tags":["23秒 · 开箱种草","9:16 竖屏","10 个镜头","商品四视图参考","Seedance 2.5"],"steps":[{"number":1,"title":"第一步：准备人物和鞋子参考图","description":"作者的流程：用 Recraft 做人物，用 GPT-2 做鞋子映射，Seedance 2.5 生成视频，ElevenLabs 配音，CapCut 剪辑。回复区给了三张图：人物脸部特写、提示词长图、一张鞋子四视图（正面、四分之三侧、俯视、正侧面）。人物特写原图收录；鞋子四视图裁成四张收录。"},{"number":2,"title":"第二步：锁定鞋子，写全局设定","description":"REFS 写明每张鞋图对应的角度，要求全片是同一双鞋：鞋头形状、带钥匙孔和冲孔的横带、皮纹、中棕色和擦色边、黑色锯齿大底；棚拍背景和灯光不要带进画面。全局设定写：左边一扇窗的上午阳光、手机视频质感、卧室橡木地板、人物服装（米白罗纹毛衣、卷边靛蓝牛仔裤、米白袜、右手两枚银戒），脸只在第 1 和第 8 镜出现。"},{"number":3,"title":"第三步：按 8 个镜头写开箱到上脚","description":"8 个镜头共 30 秒：抱盒开口 → 俯拍掀盖 → 翻开包装纸取鞋 → 侧面转鞋 → 横带微距 → 脚伸进鞋 → 俯视双脚 → 低机位看她走进画面。每镜写景别、角度、焦段、手持方式、动作和声音；只有第 1 镜有一句台词，最后要求不要字幕、不要背景音乐。"}],"references_detail":[{"id":"nastassiavideo-brown-loafers-image5-girl","number":"5","title":"@[Image 5] · 人物脸部特写","subtitle":"作者回复区第一张图原件 HQvO0dDWcAIWFHx（896x1216）· 作者说人物用 Recraft 做","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-brown-loafers-unboxing-seedance-2-5/refs/image5-girl-face.jpg","prompt":"作者未公开出图提示词（主帖写人物用 Recraft 生成）。视频提示词中的用法：REFS: @[Image 5]girl。编号对应是按提示词推断，作者没有单独标注。"},{"id":"nastassiavideo-brown-loafers-shoe1-front","number":"1","title":"乐福鞋 · 正面","subtitle":"提示词里对应 @[Image 4]front；回复区商品四视图按四等分裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-brown-loafers-unboxing-seedance-2-5/refs/shoe1-front.jpg","prompt":"作者说鞋图用 GPT-2 做了映射，没有公开出图提示词。"},{"id":"nastassiavideo-brown-loafers-shoe2-three-quarter","number":"2","title":"乐福鞋 · 四分之三侧","subtitle":"提示词里对应 @[Image 1](three quarter；回复区商品四视图按四等分裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-brown-loafers-unboxing-seedance-2-5/refs/shoe2-three-quarter.jpg","prompt":"作者说鞋图用 GPT-2 做了映射，没有公开出图提示词。"},{"id":"nastassiavideo-brown-loafers-shoe3-overhead","number":"3","title":"乐福鞋 · 俯视","subtitle":"提示词里写作 @overhead，没有编号；回复区商品四视图按四等分裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-brown-loafers-unboxing-seedance-2-5/refs/shoe3-overhead.jpg","prompt":"作者说鞋图用 GPT-2 做了映射，没有公开出图提示词。"},{"id":"nastassiavideo-brown-loafers-shoe4-profile","number":"4","title":"乐福鞋 · 正侧面","subtitle":"提示词里对应 @[Image 2]profile；回复区商品四视图按四等分裁出","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-brown-loafers-unboxing-seedance-2-5/refs/shoe4-profile.jpg","prompt":"作者说鞋图用 GPT-2 做了映射，没有公开出图提示词。"}],"storyboard":[{"number":1,"description":"0–3s 坐在床边，捧着鞋盒对镜头说第一句。"},{"number":2,"description":"3–4.3s 俯拍，戴银戒的手掀开盒盖，露出白色包装纸。"},{"number":3,"description":"4.5–6.3s 她坐在地板上，双手翻开包装纸。"},{"number":4,"description":"6.5–8.3s 手把一只鞋举起，四分之三侧面对镜头。"},{"number":5,"description":"8.5–10.8s 微距，指尖沿鞋面横带和冲孔滑过。"},{"number":6,"description":"11–12.3s 把鞋翻过来，展示黑色锯齿大底。"},{"number":7,"description":"12.5–14.8s 站着双手拿鞋，举到镜头前。"},{"number":8,"description":"15–17.3s 低机位，穿米白袜的脚伸进鞋里踩实。"},{"number":9,"description":"17.5–19.8s 站着往下拍两只鞋。"},{"number":10,"description":"20–23.4s 低机位全身，她从左边走进画面站定。"}],"constraints":"同一双鞋贯穿全片；人物服装不变；脸只在开头和结尾出现；不要字幕、不要背景音乐。与成片不符：提示词是 8 镜 30 秒，成片 23 秒、约 10 个镜头（多了坐在地板上拆纸、看鞋底、举鞋到镜头前）；提示词只有第 1 镜一句台词，成片全程有口播（作者说配音用 ElevenLabs）；提示词写不要字幕，成片有逐词弹出的大字幕，应是 CapCut 后期加的；提示词写深棕中分及肩直发，成片是及肩以下的卷发，和人物参考图一致。原文 SHOT 3 里「Tissue cracklre」是原图拼写，照录。缺口：@[Image 5] 对应脸部特写是按提示词推断，作者没有说明编号对应；口播台词没有逐句核对。","video_prompt":{"title":"Unboxing → Reveal → Wear → Walk","subtitle":"Seedance 2.5 · 9:16 · 英文完整提示词（作者回复区第二张图，逐字转录）","content":"$3b"}},{"id":"nastassiavideo-steampunk-airship-bar-seedance-2-5","title":"蒸汽朋克飞艇酒吧 · 小猴帮打 · Seedance 2.5","subtitle":"X · @nastassiavideo · Seedance 2.5（对比视频上半屏裁出）· 20秒 · 16:9","description":"飞艇酒吧里，大汉伸手抓她手腕，肩上的小猴一口咬下，她一肘反击放倒对方，回座举杯。","video":"https://media.liuyidaoai.com/tutorials/nastassiavideo-steampunk-airship-bar-seedance-2-5/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/nastassiavideo-steampunk-airship-bar-seedance-2-5/poster.jpg","duration":"20秒","durationSec":20,"styleLabel":"真人风","shots":1,"references":2,"model":"Seedance 2.5","style":"蒸汽朋克 · 黄铜吊灯顶光 · Kodak 500T 胶片颗粒 · 手持一镜到底","aspectRatio":"16/9","sourceUrl":"https://x.com/nastassiavideo/status/2092170146768921015","sourceAuthor":"@nastassiavideo","sourcePlatform":"X","sourceImpressions":883,"sourceStats":{"asOf":"2026-09-27","likes":26,"reposts":1,"bookmarks":3},"formats":["电影叙事","角色表演"],"hook":{"structure":"吧台独饮 → 大汉搭讪 → 抓手腕被猴咬 → 肘击 → 勾腿放倒 → 回座举杯","opening":"第 0 秒镜头从吧台上方往下拍她握着酒杯的手，黄铜台面反着灯光。","openingAt":0,"beats":[{"title":"冲突怎么起","text":"约 2.5–7s 平视中近景，小猴蹲在她右肩，大胡子大汉从旁边凑过来；她抬眼看他。","at":2.5},{"title":"动作段","text":"约 7.5–8s 特写他抓她手腕，小猴咬住他手指；8.5–12s 她一肘顶上他下巴；12.5–15.5s 他扑空倒下，帽子被小猴扣到眼睛上，踉跄后退。","at":7.5},{"title":"结尾怎么收","text":"约 16–20s 回到吧台正面，她坐回原位，小猴回到肩上，她举杯喝酒。","at":16}],"copyThis":"每一拍都写时间段 + 机位高度 + 动作 + 音效（SFX），并把左右手、左右肩写死（酒杯在左手、小猴在右肩、大汉在右侧），打斗方向就不会乱。","approx":true},"tags":["20秒 · 动作短剧","16:9 横屏","一镜到底","双角色设定图","Seedance 2.5 vs WAN 3.0"],"steps":[{"number":1,"title":"第一步：准备两张角色设定图","description":"作者在主帖放了两张设定图：@[Image 1] 女主 Nastassia（脸部特写 + 全身正背面，护目镜、奶白高领衬衫、棕色皮马甲、暗红裙摆、长靴），@[Image 2] 小猴（戴黄铜护目镜的绿帽、棕皮马甲、围巾，正面特写 + 正背面）。大汉只用文字描述。出图提示词没有公开。"},{"number":2,"title":"第二步：写全局设定和镜头规则","description":"光只来自吧台上方的黄铜吊灯；Kodak Vision3 500T 胶片颗粒，色彩约 50% 黄铜琥珀 + 30% 棕色皮革 + 20% 深蓝窗光；场景是飞行中的飞艇吊舱酒吧。写明三个角色的服装、声音和身高（170 / 195 / 40 cm）。镜头是一次手持运动，从吧台上方降到平视、到她下巴下方、到地板、再回平视，肘击那 0.4 秒降到 40% 速度。"},{"number":3,"title":"第三步：按 6 拍写动作和台词","description":"0–3s 坐着喝酒、大汉凑近；3–6.6s 放下杯子转头、起身；6.6–7.3s 他抓手腕、小猴咬手指；7.3–9.3s 肘击下巴；9.3–12.5s 他挥空、她勾腿放倒；12.5–20s 小猴把帽子扣他眼睛上、他仰倒、她举杯。台词两句：「Hey doll, let's play.」「Yeah. Right now.」最后写不要剪切、不要血和武器、不要字幕和音乐。作者用同一段提示词对比了 Seedance 2.5 和 WAN 3.0。"}],"references_detail":[{"id":"nastassiavideo-steampunk-image1-nastassia","number":"1","title":"@[Image 1] · 女主 Nastassia 设定图","subtitle":"主帖第一张图原件 HQjhl7hXwAAMC6R（1536x1024）· AI 角色设定图","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-steampunk-airship-bar-seedance-2-5/refs/image1-nastassia-steampunk-sheet.jpg","prompt":"作者未公开出图提示词。视频提示词中的用法：@[Image 1]= Nastassia, controls face, hair and wardrobe only, full close up fidelity in every framing."},{"id":"nastassiavideo-steampunk-image2-monkey","number":"2","title":"@[Image 2] · 小猴设定图","subtitle":"主帖第二张图原件 HQjhl7VXUAALb3z（1536x1024）· AI 角色设定图","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-steampunk-airship-bar-seedance-2-5/refs/image2-monkey-sheet.jpg","prompt":"作者未公开出图提示词。视频提示词中的用法：@[Image 2]= the Monkey, controls face, fur and outfit only."}],"storyboard":[{"number":1,"description":"0–2s 吧台上方俯拍，她握着酒杯的手。"},{"number":2,"description":"2.5–7s 平视，小猴在她右肩，大汉从旁边凑近，她抬眼看他。"},{"number":3,"description":"7.5–8s 特写：他抓她手腕，小猴咬住他手指。"},{"number":4,"description":"8.5–12s 她一肘顶上他下巴，他头往后仰。"},{"number":5,"description":"12.5–14.5s 快速晃动，他倒向地面，小猴抢过帽子。"},{"number":6,"description":"15–16s 帽子扣在他眼睛上，他踉跄后退。"},{"number":7,"description":"16–20s 吧台正面，她坐回原位，小猴回到肩上，她举杯喝酒。"}],"constraints":"角色脸和服装全程一致；酒杯在左手、小猴在右肩、大汉在她右侧；一镜到底不剪切；不要血和武器，不要字幕、水印、Logo 和音乐。来源说明：原帖视频是 1080x1440 上下分屏对比，中间有「Seedance 2.5 /Wan 3.0」标签条；上半屏 Seedance 2.5 在 0–20.07s 播放（下半屏静止），下半屏 WAN 3.0 在 20.07–40.1s 播放。本页成片只裁出上半屏 0–20.07s（1080x606，去掉标签条和黑边），音频是原视频同一时段音轨，未改动。与成片不符：成片里她大部分时间坐着，提示词写的「起身、凳子被踢翻」看不清；中段有几次快速甩镜，是否有硬切没有逐帧确认。缺口：两张设定图的出图提示词没公开；台词没有逐句核对。","video_prompt":{"title":"Steampunk Airship Bar · Nastassia & the Monkey","subtitle":"Seedance 2.5（与 WAN 3.0 共用同一段提示词）· 英文完整提示词（主帖第三张图，逐字转录）","content":"$3c"}},{"id":"nastassiavideo-corset-hangar-vhs-seedance-2-0","title":"黑色束腰机库大片 · VHS 复古手持 · Seedance 2.0","subtitle":"X · @nastassiavideo · Seedance 2.0 · 24秒 · 16:9 · 提示词为反推","description":"黑色皮革束腰 × 私人飞机机库：手持 VHS 摄像机质感的复古时装短片，带橙色时间码。","video":"https://media.liuyidaoai.com/tutorials/nastassiavideo-corset-hangar-vhs-seedance-2-0/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/nastassiavideo-corset-hangar-vhs-seedance-2-0/poster.jpg","duration":"24秒","durationSec":24,"styleLabel":"真人风","shots":14,"references":5,"model":"Seedance 2.0","style":"机库逆光 · 青橙调 · 90 年代摄像机 VHS 颗粒 · 橙色时间码","aspectRatio":"16/9","sourceUrl":"https://x.com/nastassiavideo/status/2090778378701832651","sourceAuthor":"@nastassiavideo","sourcePlatform":"X","sourceImpressions":731,"sourceStats":{"asOf":"2026-09-27","likes":16,"reposts":0,"bookmarks":4},"formats":["时尚大片","产品广告"],"hook":{"structure":"背影走向飞机 → 逆光特写 → 机翼剪影 → 局部特写串联 → 奔跑甩镜 → 机库大门剪影收尾","opening":"第 0 秒低机位从背后拍她走过机库、走向白色私人飞机，回头看镜头；右下角是橙色摄像机时间码「PM 2:29」。","openingAt":0,"beats":[{"title":"逆光和剪影","text":"约 2–4s 靠在机身上撩头发，身后强烈暖色逆光；4–5.5s 超低机位，她站在机翼上举起双臂，顶棚天窗打下光柱。","at":2},{"title":"局部特写串联","text":"约 5.5–13.8s 连续特写：手穿过发丝、一道光条横过眼睛、脖颈和金链、坐在舷梯扶手上、手指拉紧束腰上的两个银扣。","at":5.5},{"title":"节奏和收尾","text":"约 14–16s 她跑过机库，甩镜加运动模糊转场；之后正面走来、嘴唇特写、束腰特写、站在机头旁；21.5–23.6s 两台发动机之间，她的剪影站在敞开的机库大门口。","at":14}],"copyThis":"作者公开的复古质感关键词：handheld camera, shaky cam, VHS style, 90s camcorder, timestamp, motion blur。再把服装单独做成一张商品图当参考，锁定束腰的扣子和轮廓。","approx":true},"tags":["24秒 · 时装短片","16:9 横屏","VHS 复古手持","提示词为反推","角色一致性测试"],"steps":[{"number":1,"title":"第一步：作者公开的流程和图片","description":"作者写的流程：用 Recraft 做人物，用 Seedream 4.5 出了几张写真，用 Seedance 2.0 生成视频测角色一致性，CapCut 剪辑。主帖附了三张图：束腰商品图、人物脸部特写、坐在飞机舷梯上的定妆照。三张原图都已收录，出图提示词作者没公开，本页按图反推（见参考图说明和 PROMPT_REVERSED.txt）。作者没有公开视频提示词，只公开了复古质感关键词。"},{"number":2,"title":"第二步：作者公开的复古关键词","description":"作者原文：handheld camera, shaky cam, VHS style, 90s camcorder, timestamp, motion blur。成片里的对应效果：全程手持轻晃、VHS 颗粒和色边、右下角橙色时间码（远景 PM 2:29，特写 PM 2:34）、奔跑和转场时的运动模糊。"},{"number":3,"title":"第三步：按成片反推的分镜写法（作者未公开，此为按成片反推）","description":"作者未公开，此为按成片反推。按 2fps 逐帧看成片，拆成 14 个镜头并写时间码：场景是午后的飞机机库，天窗和敞开的大门打进低角度硬光，有烟雾和光柱；青橙调色；每镜写景别、机位、动作和光线。成片应是多段生成后在 CapCut 剪在一起的，分段方式也是推测。反推的英文提示词见下方视频提示词。"}],"references_detail":[{"id":"nastassiavideo-corset-hangar-image1-corset","number":"1","title":"黑色皮革束腰 · 商品图","subtitle":"作者主帖第一张图原件","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-corset-hangar-vhs-seedance-2-0/refs/image1-corset-HQPvdE2XEAAMf57.jpg","prompt":"作者未公开，此为按成片反推。（本条按作者原图反推出图提示词）Studio product photo of a black high-gloss leather corset top shown on an invisible mannequin, front view, centred on a seamless white background. Sculpted rounded cups with a deep V between them, front cord lacing visible in the gap, two horizontal leather belts with polished silver pin buckles across the ribs, curved panel seams, flared peplum hips, loose lace ends hanging below the hem. Soft even softbox lighting with long specular highlights on the leather, gentle floor shadow, e-commerce clarity, 3:4."},{"id":"nastassiavideo-corset-hangar-image2-face","number":"2","title":"人物脸部特写 · 作者原图","subtitle":"作者主帖第二张图原件 HQPvdE9WcAAmodJ（896x1216）· 作者说人物用 Recraft 做","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-corset-hangar-vhs-seedance-2-0/refs/image2-face-HQPvdE9WcAAmodJ.jpg","prompt":"作者未公开，此为按成片反推。（本条按作者原图反推出图提示词）Editorial beauty close-up portrait of a young adult woman, face tilted slightly, voluminous tousled golden-blonde curls with wispy bangs, hazel-green eyes with defined lashes, light freckles, dewy skin with visible pores, glossy nude lips slightly parted. Hard directional daylight from the upper left casting a deep shadow over one side of the face, off-white background, shot on 85mm, shallow depth of field, fashion magazine realism, 3:4."},{"id":"nastassiavideo-corset-hangar-image3-jet-stairs","number":"3","title":"飞机舷梯定妆照 · 作者原图","subtitle":"作者主帖第三张图原件 HQPvdEzWoAA2EQ6（1290x1717）· 作者说写真用 Seedream 4.5 做","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-corset-hangar-vhs-seedance-2-0/refs/image3-jet-stairs-HQPvdEzWoAA2EQ6.jpg","prompt":"作者未公开，此为按成片反推。（本条按作者原图反推出图提示词）Fashion editorial photo, low angle: the same curly blonde woman (@Image2) sits on the steel roll-in stairs of a white private jet with a navy stripe inside a hangar, forearm resting on her knee, cool direct gaze. She wears the black leather corset (@Image1) over a black glossy bodysuit and black high-gloss thigh-high stiletto boots. Polished handrails frame her, haze and warm backlight glow from the left, hangar windows behind, cinematic contrast, 3:4."},{"id":"nastassiavideo-corset-hangar-frame-13s","number":"4","title":"束腰银扣特写 · 成片截帧","subtitle":"成片第 13 秒截帧，不是作者的原参考图","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-corset-hangar-vhs-seedance-2-0/refs/frame-13s-corset-buckles.jpg","prompt":"作者未公开，此为按成片反推。（成片截帧，不是原参考图）Extreme close-up of the black leather corset belts with two polished silver pin buckles, a hand with dark burgundy nails tugging the lower strap, shallow depth of field, warm side light, VHS grain, orange camcorder timestamp \"PM 2:34\" bottom right, 16:9."},{"id":"nastassiavideo-corset-hangar-frame-22s","number":"5","title":"机库大门剪影 · 成片截帧","subtitle":"成片第 22.5 秒截帧，不是作者的原参考图","image":"https://media.liuyidaoai.com/tutorials/nastassiavideo-corset-hangar-vhs-seedance-2-0/refs/frame-22s-hangar-door-silhouette.jpg","prompt":"作者未公开，此为按成片反推。（成片截帧，不是原参考图）Wide symmetrical shot from deep inside a dark aircraft hangar, two jet engine nacelles as dark shapes in the foreground left and right, a small silhouette of a woman standing in the bright open hangar door, warm sunset sky and haze outside, industrial windows on the door panels, VHS grain, orange camcorder timestamp \"PM 2:29\" bottom right, 16:9."}],"storyboard":[{"number":1,"description":"0–2s 低机位背影，她走向白色私人飞机，回头看镜头。"},{"number":2,"description":"2–4s 靠在机身上撩头发，暖色逆光光晕。"},{"number":3,"description":"4–5.5s 超低机位，站在机翼上举起双臂，天窗光柱。"},{"number":4,"description":"5.5–7s 手穿过逆光的发丝。"},{"number":5,"description":"7–8.5s 眼睛大特写，一道光条横过双眼。"},{"number":6,"description":"8.5–9.5s 脖颈、肩膀和金链特写。"},{"number":7,"description":"9.5–11.5s 低机位，坐在飞机舷梯上，手臂搭在扶手上。"},{"number":8,"description":"11.5–13.8s 手指拉紧束腰腰带，两个银扣占满画面。"},{"number":9,"description":"14–15.8s 侧面跟拍她跑过机库，甩镜运动模糊。"},{"number":10,"description":"16–17s 正面，她朝镜头走来。"},{"number":11,"description":"17–18s 嘴唇大特写。"},{"number":12,"description":"18–20s 束腰正面特写，双手抚平腰部。"},{"number":13,"description":"20–21.5s 全身，站在机头旁。"},{"number":14,"description":"21.5–23.6s 两台发动机之间，她的剪影站在敞开的机库大门口，保持到结尾。"}],"constraints":"作者未公开视频提示词；本页视频提示词、分镜时间码和出图提示词均为按成片反推，不是作者原文。作者原文只有一句复古关键词：handheld camera, shaky cam, VHS style, 90s camcorder, timestamp, motion blur。成片截帧（第 13 秒、第 22.5 秒）不是作者的原参考图。缺口：音轨是持续配乐、没有对白，具体曲目无法确认；是否分多段生成无法确认；评论区检索没有找到作者的补充说明。","video_prompt":{"title":"Corset × Hangar · Retro VHS (Reverse-engineered)","subtitle":"作者未公开，此为按成片反推 · 按作者标注的 Seedance 2.0 写 · 16:9","content":"$3d"}},{"id":"filmera-pvz-sunflowers-redemption-seedance-2-5","title":"植物大战僵尸 · 向日葵的救赎 · Seedance 2.5","subtitle":"Filmera 模板 · Seedance 2.5 + GPT Image 2.5 · 30秒 · 16:9","description":"植物大战僵尸真人版：戴锅的 Dave 挡在豌豆射手前护住黑化的向日葵，一枪阳光后重回花园。","video":"https://media.liuyidaoai.com/tutorials/filmera-pvz-sunflowers-redemption-seedance-2-5/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/filmera-pvz-sunflowers-redemption-seedance-2-5/poster.jpg","duration":"30秒","durationSec":30,"styleLabel":"真人风","shots":14,"references":8,"model":"Seedance 2.5（bytedance/seedance-2.5，480p，生成音频）+ GPT Image 2.5 Flare（参考图）","style":"游戏 IP 真人化 · 原片直出感 · 低对比低饱和 · 剧烈手持晃动","aspectRatio":"16/9","sourceUrl":"https://www.filmera.ai/templates/f963cb32-3319-432d-93e6-ce858f1e2342","sourceAuthor":"Filmera 模板（未署名创作者）","sourcePlatform":"Filmera","sourceImpressions":155,"sourceStats":{"asOf":"2026-09-27"},"formats":["电影叙事","角色表演"],"hook":{"structure":"门廊砸僵尸 → 冲进后院 → 张开双臂护住黑化向日葵 → 被扑倒、枪口顶下巴 → 阳光一枪 → 回到往日花园","opening":"第 0 秒机位在 Dave 右肩后，僵尸拖着步子走到拱形前门，Dave 用猎枪枪托砸过去，僵尸仰面摔下台阶。","openingAt":0,"beats":[{"title":"冲突怎么起","text":"约 4–5s 跟着他穿过昏暗走廊推开后门；6–7s 后院篱笆边，黑化向日葵转头，两眼发白光，两侧豌豆射手转向它；8–10s 他跑到中间张开双臂挡住。","at":4},{"title":"反转","text":"约 11–17s 低机位近景，他求豌豆射手别开火，又端起猎枪对准它们；18–19s 黑化向日葵从背后起身把他扑倒；20s 插入一段金色向日葵的回忆。","at":11},{"title":"结尾怎么收","text":"约 23–26s 向日葵用叶子举枪顶住他下巴，白光眼慢慢变成暖琥珀色；27–28s 枪口喷出金色阳光，他在光里散成光点；29–30s 回到阳光后院，他把脸贴在笑着的向日葵上。","at":23}],"copyThis":"先给每个角色、道具、场景各出一张设定图，再在视频提示词开头用 @Image1–8 逐个绑定；情绪只写成看得见的动作（喘气、吞口水、手抖），不写「害怕」「难过」这类词。","approx":true},"tags":["30秒 · 短片","16:9 横屏","14 个镜头","8 张参考图","Seedance 2.5"],"steps":[{"number":1,"title":"第一步：用 GPT Image 2.5 出 8 张参考图","description":"模板分三组：角色（Dave、僵尸）、道具（正常向日葵、黑化向日葵、豌豆射手、双管猎枪）、场景（前门廊、后院空景）。都用 openai/gpt-image-2.5-flare、3:2 出图；设定图统一写「浅奶油色背景、多视角、英文标签」，并要求像未修图的真实照片。每张的完整出图提示词见下方参考图。"},{"number":2,"title":"第二步：视频提示词先写全局规则","description":"[GLOBAL] 段写：原片直出感、低对比低饱和、不调色、剧烈手持晃动；逐个写出 8 个对象的外观并标「match @ImageN exactly」；表演要像真人（呼吸、眨眼、吞咽、手收紧），情绪只写动作；英文台词对口型；不要字幕、不要背景音乐，只要现场声。"},{"number":3,"title":"第三步：写 14 个镜头，一次生成 30 秒","description":"Shot 1–14 每条写景别、机位、动作和音效（CRACK、THUD、CLICK-CLACK、HUM），台词三句：「Please… don't hurt her.」「Don't hurt her. Trust me… she'll turn back.」「No…」。在 Seedance 2.5 里选 16:9、30 秒、480p、打开生成音频。模板里同一提示词跑了 3 条，本页用的是模板封面那条。"}],"references_detail":[{"id":"filmera-pvz-sunflower-image1-dave","number":"1","title":"@Image1 · Dave 人物设定图","subtitle":"人物：姜色大胡子、倒扣铁锅当头盔、白 Polo 衫、卡其裤；模板节点「01 · DAVE」，openai/gpt-image-2.5-flare 3:2","image":"https://media.liuyidaoai.com/tutorials/filmera-pvz-sunflowers-redemption-seedance-2-5/refs/01-dave.jpg","prompt":"$3e"},{"id":"filmera-pvz-sunflower-image2-sunflower","number":"2","title":"@Image2 · 向日葵（正常）","subtitle":"种子盘天然长成笑脸的真向日葵；模板节点「02 · SUNFLOWER」，openai/gpt-image-2.5-flare 3:2","image":"https://media.liuyidaoai.com/tutorials/filmera-pvz-sunflowers-redemption-seedance-2-5/refs/02-sunflower.jpg","prompt":"Creature reference sheet on a clean light cream background, the same living plant shown four ways, labeled: a FULL VIEW, a THREE-QUARTER VIEW, a CLOSE-UP of the face, and a CLOSE-UP of the stem and leaves. The creature: a real sunflower as tall as a man's chest on a thick fuzzy green stem with broad heart-shaped leaves; a large head ringed with bright golden petals; the dark brown seed disc naturally forms a gentle face: two small arcs of darker seeds as smiling closed eyes, a soft curved line of seeds as a smile, a faint warm blush of orange seeds on the cheeks. Real plant textures, fine hairs on the stem, pollen dust on the petals. Every panel is a REAL photograph of the same plant — soft natural daylight, slightly low contrast and low saturation, no retouching, unedited photo look. Clean minimal English labels only, no other text."},{"id":"filmera-pvz-sunflower-image3-sunflower-turned","number":"3","title":"@Image3 · 黑化向日葵","subtitle":"紫黑卷曲花瓣、裂口獠牙、白光眼；模板节点「03 · SUNFLOWER TURNED」，openai/gpt-image-2.5-flare 3:2","image":"https://media.liuyidaoai.com/tutorials/filmera-pvz-sunflowers-redemption-seedance-2-5/refs/03-sunflower-turned.jpg","prompt":"Creature reference sheet on a clean light cream background, the same living plant shown four ways, labeled: a FULL VIEW, a THREE-QUARTER VIEW, a CLOSE-UP of the face with the mouth open, and a CLOSE-UP of a leaf curled tight around a stick. The creature: a tall sunflower that has turned, with the same stem, leaf shape and head size as a healthy sunflower as tall as a man's chest: its petals now dark purple-black, curled, torn and leathery; the seed disc split across into a wide mouth ringed with sharp seed-shell fangs and a dark wet throat; two deep pits in the disc glowing solid white as eyes; the stem darkened with purple veins and grey patches of rot; the broad leaves thickened, curled and strong enough to grip. Practical-effects creature look, wet organic surfaces. Every panel is a REAL photograph of the same plant — soft natural daylight, slightly low contrast and low saturation, no retouching, unedited photo look. Clean minimal English labels only, no other text."},{"id":"filmera-pvz-sunflower-image4-pea-plant","number":"4","title":"@Image4 · 豌豆射手","subtitle":"齐腰高、木质炮口的真实植物；模板节点「04 · PEA PLANT」，openai/gpt-image-2.5-flare 3:2","image":"https://media.liuyidaoai.com/tutorials/filmera-pvz-sunflowers-redemption-seedance-2-5/refs/04-pea-plant.jpg","prompt":"Creature reference sheet on a clean light cream background, the same living plant shown four ways, labeled: a FULL VIEW planted in dark soil, a SIDE VIEW with the head reared back to fire, a CLOSE-UP of the muzzle, and a CLOSE-UP of the stem base and roots. The creature: a waist-high plant with a thick fibrous green stem the width of a forearm, veined and glistening with sap; the head is a bulbous green pod the size of a football with a single round hollow wooden muzzle at the front, its lips ringed with tiny fibres; two ragged leaves at the base; a smooth green pea the size of a golf ball rests in the muzzle. Practical-effects creature look, wet organic surfaces, real soil. Every panel is a REAL photograph of the same plant — soft natural daylight, slightly low contrast and low saturation, no retouching, unedited photo look. Clean minimal English labels only, no other text."},{"id":"filmera-pvz-sunflower-image5-zombie","number":"5","title":"@Image5 · 僵尸设定图","subtitle":"灰绿皮肤、棕西装、红条纹领带；模板节点「05 · ZOMBIE」，openai/gpt-image-2.5-flare 3:2","image":"https://media.liuyidaoai.com/tutorials/filmera-pvz-sunflowers-redemption-seedance-2-5/refs/05-zombie.jpg","prompt":"$3f"},{"id":"filmera-pvz-sunflower-image6-shotgun","number":"6","title":"@Image6 · 双管猎枪","subtitle":"外露双击锤、胡桃木枪托；模板节点「06 · SHOTGUN」，openai/gpt-image-2.5-flare 3:2","image":"https://media.liuyidaoai.com/tutorials/filmera-pvz-sunflowers-redemption-seedance-2-5/refs/06-shotgun.jpg","prompt":"Prop reference sheet on a clean light cream background, the same object shown three ways, labeled: a SIDE VIEW, a THREE-QUARTER VIEW, and a CLOSE-UP of the twin muzzles and hammers. The object: an old side-by-side double-barrel shotgun with twin exposed hammers, a scuffed walnut stock and fore-end, worn blued steel barrels with patches of rust. Every panel is a REAL photograph of the same prop — soft natural daylight, slightly low contrast and low saturation, no retouching, unedited photo look. Clean minimal English labels only, no other text."},{"id":"filmera-pvz-sunflower-image7-front-porch","number":"7","title":"@Image7 · 前门廊空景","subtitle":"灰色护墙板房子、拱形木门；模板节点「07 · FRONT PORCH」，openai/gpt-image-2.5-flare 3:2","image":"https://media.liuyidaoai.com/tutorials/filmera-pvz-sunflowers-redemption-seedance-2-5/refs/07-front-porch.jpg","prompt":"Empty scene plate, no people, no creatures, no text, no logos: the front porch of a pale grey clapboard suburban house, a dark wooden arched front door with small glass panes and black iron hinges, worn painted porch boards, three wooden steps down to a stone path, a potted shrub beside the door, low late-afternoon sun raking across the siding. Photoreal, natural light only, slightly low contrast and low saturation, unedited photo look. No other text."},{"id":"filmera-pvz-sunflower-image8-back-yard","number":"8","title":"@Image8 · 后院空景","subtitle":"草坪、白色尖桩篱笆、洒水壶；模板节点「08 · BACK YARD」，openai/gpt-image-2.5-flare 3:2","image":"https://media.liuyidaoai.com/tutorials/filmera-pvz-sunflowers-redemption-seedance-2-5/refs/08-back-yard.jpg","prompt":"Empty scene plate, no people, no creatures, no text, no logos: the back yard of the same pale grey clapboard house seen from its arched back door, a lush green lawn in low warm afternoon sun, a white picket fence along the far side with neighbouring rooftops behind it, dark soil garden beds along the fence and on both sides of the lawn, a green watering can on the grass, the house wall with white window frames at the left edge. Photoreal, natural light only, slightly low contrast and low saturation, unedited photo look. No other text."}],"storyboard":[{"number":1,"description":"0–3.5s 门廊：僵尸走到前门，Dave 用枪托砸倒它，转身进屋。"},{"number":2,"description":"4–5.5s 跟拍穿过昏暗走廊，推开后门，阳光涌进来。"},{"number":3,"description":"6–7s 后院全景：篱笆边的黑化向日葵转头，两眼白光。"},{"number":4,"description":"8–10s 他跑上草坪，站在豌豆射手和向日葵之间张开双臂。"},{"number":5,"description":"11–14s 低机位近景，前景是模糊的豌豆射手，他喘着气求它们别开火。"},{"number":6,"description":"15s 豌豆射手特写，炮口对准。"},{"number":7,"description":"16–17s 他端起猎枪对准豌豆射手，双手发抖。"},{"number":8,"description":"18–19s 黑化向日葵从背后起身把他扑倒，獠牙和白光眼压到镜头上方。"},{"number":9,"description":"20s 回忆：阳光下金色的向日葵。"},{"number":10,"description":"21–22s 俯拍他仰躺在草地上，锅歪在头上。"},{"number":11,"description":"23–24s 贴地侧拍：向日葵用叶子举枪顶住他下巴。"},{"number":12,"description":"25–26s 向日葵脸部特写，白光眼变成暖琥珀色。"},{"number":13,"description":"27–28s 枪口喷出金色阳光，他在光里大喊、散成光点。"},{"number":14,"description":"29–30s 往日的后院，他把脸贴在笑着的向日葵上。"}],"constraints":"8 个对象都要和各自参考图一致；原片直出感、不调色、剧烈手持；情绪只写成动作；英文台词对口型；不要字幕、不要背景音乐。与提示词对照：第 14 镜提示词写他穿干净白衬衫、手拿洒水壶跪在花旁，成片里是他俯身把脸贴在花上，没看到洒水壶，是否跪着看不清；第 12 镜「一片花瓣变回金色」在成片里不明显。版本说明：模板 GEN 1 同一提示词有 3 条 take（都是 854x480、30 秒），节点标记的 best_take 是第 1 条，模板封面预览与第 3 条同片；本页用第 3 条，因为它更贴近提示词后半段（黑化向日葵举枪、眼睛变色、人散成光点），第 1 条后半段直接变成金色向日葵举枪。存疑：8 张出图时作者还附了另一段视频的截帧作为输入参考，这些截帧未在模板里单独公开，本页没有收录；模板没有署名创作者。","video_prompt":{"title":"Sunflower's Redemption · GEN 1","subtitle":"Seedance 2.5 · 16:9 · 30s · 480p · 生成音频 · Filmera 模板页原文","content":"$40"}},{"id":"naiknelofar788-me-time-dark-circles-seedance","title":"报复性熬夜 · 黑眼圈越刷越深 · Seedance 2.5","subtitle":"X · @Naiknelofar788 · Seedance 2.5 · 30秒 · 16:9","description":"Seedance 2.5 熬夜刷手机小短剧：下班后舍不得睡，黑眼圈随时间一路加深，最后黑屏字幕收梗。","video":"https://media.liuyidaoai.com/tutorials/naiknelofar788-me-time-dark-circles-seedance/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/naiknelofar788-me-time-dark-circles-seedance/poster.jpg","duration":"30秒","durationSec":30,"styleLabel":"写实生活","shots":5,"references":0,"model":"Seedance 2.5","style":"写实生活喜剧 · 暖色卧室转冷色手机光 · 网络梗短剧","aspectRatio":"16/9","sourceUrl":"https://x.com/Naiknelofar788/status/2103770487117713631","sourceAuthor":"@Naiknelofar788","sourcePlatform":"X","sourceImpressions":1846,"sourceStats":{"asOf":"2026-09-27","likes":51,"reposts":0,"bookmarks":30},"formats":["角色表演","电影叙事"],"hook":{"structure":"下班瘫倒 → 手机光下黑眼圈加深 → 照镜子又解锁 → 黑屏字幕收梗","opening":"开场是暖黄台灯的卧室，女孩下班进门、跪上床再整个人趴倒——先用一个人人都有过的「终于到家」动作把人拉进来。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"约 4.6s 切到黑暗里被手机照亮的脸，她笑着刷手机，屏上字幕「Me: I should sleep early tonight.」；机位几乎不动，约 13–20s 黑眼圈一点点加深成墨色，字幕换成「But I didn't get enough ME TIME today.」。","at":4.6},{"title":"照镜子又解锁","text":"约 21s 前景举着亮屏手机，后面暗色镜面里是满眼黑眼圈的自己；她看看自己、又低头把手机点亮继续刷。","at":21},{"title":"结尾怎么收","text":"约 26s 贴脸大特写，只剩被屏幕照亮的黑眼圈；约 28.8s 切黑屏，白字「Tomorrow's problem.」和一行小字 Me time / Sleep / Regrets: loading…。","at":28.8}],"copyThis":"同一个机位、同一张被手机照亮的脸，只让黑眼圈一路加深来表现时间流逝，最后用黑屏字幕把梗说破。","approx":true},"tags":["30秒 · 生活喜剧","16:9 横屏","Seedance 2.5","纯文生视频","熬夜刷手机","黑屏字幕收梗"],"steps":[{"number":1,"title":"先用一句话定题，再分 5 场写剧情","description":"提示词开头一句定调：一个下班后终于有自己时间、所以不肯睡觉的年轻女孩，幽默写实生活短片。然后按 Scene 1–5 写：下班瘫倒 → 「Finally, Me Time」刷手机 → 时间小偷（时间跳跃）→ 视觉比喻（照镜子后又解锁手机）→ 笑点收尾。纯文生视频，没有参考图。"},{"number":2,"title":"把屏幕字幕和时间跳字原样写进提示词","description":"要出现的字都用引号写死：「Me: I should sleep early tonight.」「But I didn't get enough ME TIME today.」、结尾黑屏「Tomorrow's problem.」和三行小字。时间写成 11:47 PM → 12:38 AM → 1:52 AM → 3:07 AM，并要求每跳一次黑眼圈更深、手机始终明亮。注意：成片里时间数字没有清晰渲染，结尾小字也出现了错字。"},{"number":3,"title":"最后写风格、画幅和「黑眼圈变钟影」的小花招","description":"Style 段写暖色卧室光混冷色手机光、夸张但不过火的黑眼圈、表情表演、浅景深和少量手持感，并写了 9:16、15–20 秒；最后的 Extra visual trick 要求黑眼圈逐渐像时钟阴影。实际成片是 16:9、30 秒，黑眼圈只是一路加深，没有出现时钟形状——跟做时画幅和时长要在平台参数里单独设。"}],"references_detail":[],"storyboard":[{"number":1,"description":"约 0–4.6s 暖黄卧室：女孩下班进门，跪上床后趴倒，放松地翻身。"},{"number":2,"description":"约 4.6–13s 黑暗中手机照亮的脸，她笑着刷手机，字幕「Me: I should sleep early tonight.」。"},{"number":3,"description":"约 13–21s 同一机位，黑眼圈逐渐加深成墨色，字幕换成「But I didn't get enough ME TIME today.」。"},{"number":4,"description":"约 21–26s 前景亮屏手机，后方暗色镜面里满眼黑眼圈的自己，她又把手机点亮继续刷。"},{"number":5,"description":"约 26–30s 贴脸大特写后切黑屏，白字「Tomorrow's problem.」与小字 Me time / Sleep / Regrets: loading…。"}],"constraints":"写实生活喜剧质感，暖色卧室光与冷色手机光对比；黑眼圈随时间逐步加深但保持夸张得体；屏幕字幕与结尾黑屏文字按引号原文；表情表演为主，浅景深，少量手持。缺口：纯文生视频无参考图；提示词写 9:16、15–20 秒，成片为 16:9、30 秒；时间跳字未清晰出现，『手机变大人变小』与『黑眼圈变时钟阴影』未在成片中呈现；结尾小字渲染为『Me l time』；音频（耳语台词）未转录。","video_prompt":{"title":"熬夜刷手机 · Seedance 2.5 提示词","subtitle":"Scene 1–5 · Seedance 2.5 · 纯文生视频 · 英文完整提示词（主帖原文）","content":"$41"}},{"id":"shamsamin56-tharog-hunter-cave-chase-seedance","title":"别惹 THAROG · 猎人射兽逃进山洞 · Seedance 2.5","subtitle":"X · @ShamsAmin56 · Seedance 2.5 · videoduck · 30秒 · 16:9","description":"Seedance 2.5 写实奇幻追逐：猎人一箭射中巨兽 THAROG 后被一路狂追，扑进山洞，巨兽撞上洞口。","video":"https://media.liuyidaoai.com/tutorials/shamsamin56-tharog-hunter-cave-chase-seedance/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/shamsamin56-tharog-hunter-cave-chase-seedance/poster.jpg","duration":"30秒","durationSec":30,"styleLabel":"史诗写实","shots":13,"references":2,"model":"Seedance 2.5（videoduck）","style":"写实奇幻 · 荒漠峡谷猎人与巨兽追逐 · 电影感运镜","aspectRatio":"16/9","sourceUrl":"https://x.com/ShamsAmin56/status/2103849845467537562","sourceAuthor":"@ShamsAmin56","sourcePlatform":"X","sourceImpressions":15070,"sourceStats":{"asOf":"2026-09-27","likes":201,"reposts":24,"bookmarks":68},"formats":["电影叙事"],"hook":{"structure":"潜伏拉弓 → 射中怒吼 → 追逐 → 逃进山洞","opening":"第 0 秒猎人蹲在前景巨石后，远处荒漠里一头长着巨角的怪兽正站着——一眼就知道有人要惹大麻烦。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"约 1–6s 从猎人脸推到抽箭、搭弦、越肩瞄准、满弓手部特写；约 6–8s 松弦，镜头跟着箭飞过峡谷，巨兽越来越大；约 8.5s 箭扎进它眼睛上方。","at":1},{"title":"怒吼与追逐","text":"约 10s 低角度正面怒吼，扬起尘土；约 12–14s 它发现猎人，猎人转身狂奔；约 15–22s 侧跟、低机位、俯拍轮番切，巨兽撞断枯树越追越近，山洞在崖壁上出现。","at":10},{"title":"结尾怎么收","text":"约 23–26s 从洞内往外拍，猎人扑进洞口翻滚，巨兽冲到洞口撞上岩壁、扬尘糊满画面；约 28.5s 洞内逆光剪影，猎人喘气，巨兽被卡在明亮的洞口外。","at":23}],"copyThis":"提示词按每 1 秒写一格，并规定每一秒都从上一秒的位置和动作接着来——30 秒追逐的空间方向和箭伤位置才能一直连得上。","approx":true},"tags":["30秒 · 奇幻追逐","16:9 横屏","Seedance 2.5","videoduck","双角色设定图","逐秒时间轴"],"steps":[{"number":1,"title":"先准备猎人和巨兽两张设定图，开头锁死参考","description":"作者在提示词回复里附了两张设定图：猎人（多角度全身、脸部特写、弓箭细节）和 THAROG（侧/正/背/俯视图、身高对比、头角与爪细节）。提示词开头 STRICT REFERENCE LOCK 要求两者外观完全照参考，并提前规定箭射中后一直插在同一只眼上。注意：提示词里写 HUNTER=image2、THAROG=image1，和 X 帖的发图顺序相反，上传时按图内容对应。"},{"number":2,"title":"每 1 秒写一格：机位 + 动作 + 转场","description":"00:00–00:30 每一秒都写 CAMERA（焦段、机位）、动作和 TRANSITION（这一秒怎么接到下一秒），例如松弦时镜头跟箭飞、冲进洞口前镜头先进洞里回拍。整条片几乎不用无关硬切，靠运镜和扬尘、碎木、灰尘把镜头接起来。"},{"number":3,"title":"最后写连续性规则，防止位置和伤势重置","description":"CONTINUITY RULE 把 30 秒当成一个连续事件：每一秒继承上一秒的位置、方向、动作惯性、镜头轨迹、扬尘、箭伤位置和角的损伤；山洞必须先在远景出现、追逐中逐渐看清，最后才成为逃生口；角只在 27–28 秒断裂。"}],"references_detail":[{"id":"ref-shamsamin56-hunter","number":"1","title":"猎人角色设定图","subtitle":"作者提示词回复帖第 1 张 · 对应 [REFERENCE 1 — HUNTER]","image":"https://media.liuyidaoai.com/tutorials/shamsamin56-tharog-hunter-cave-chase-seedance/refs/ref-hunter.jpg","prompt":"作者未公开这张设定图的出图提示词。视频提示词中的用法：[REFERENCE 1 — HUNTER] Use the supplied Hunter reference exactly: same face, hair, beard, physique, fur-and-leather clothing, bow, arrows, quiver and materials.（提示词里标为 image2，与 X 帖发图顺序相反。）"},{"id":"ref-shamsamin56-tharog","number":"2","title":"THAROG 巨兽设定图","subtitle":"作者提示词回复帖第 2 张 · 对应 [REFERENCE 2 — THAROG]","image":"https://media.liuyidaoai.com/tutorials/shamsamin56-tharog-hunter-cave-chase-seedance/refs/ref-tharog.jpg","prompt":"作者未公开这张设定图的出图提示词。视频提示词中的用法：[REFERENCE 2 — THAROG] Use the supplied THAROG reference exactly: same giant proportions, skull, crescent horns, armored dorsal plates, fur, hide coloration, limbs and tail.（提示词里标为 image1，与 X 帖发图顺序相反。）"}],"storyboard":[{"number":1,"description":"约 0–1s 猎人蹲在前景巨石后，远处巨兽站在荒漠中。"},{"number":2,"description":"约 1–4s 推到猎人脸，抽箭、搭弦。"},{"number":3,"description":"约 4–6s 越肩瞄准，满弓手部特写。"},{"number":4,"description":"约 6–8s 松弦，镜头跟随箭飞过峡谷，巨兽迎面变大。"},{"number":5,"description":"约 8–10s 慢动作特写：箭扎进眼睛上方，头猛地后仰。"},{"number":6,"description":"约 10–12s 低角度正面，巨兽仰头怒吼，尘土震落。"},{"number":7,"description":"约 12–14s 独眼搜寻，发现巨石旁的猎人，猎人转身逃跑。"},{"number":8,"description":"约 14–18s 巨兽冲锋，侧跟与低机位追逐，距离拉近。"},{"number":9,"description":"约 18–20s 俯拍/正面跟拍，巨兽撞断枯树，崖壁上的洞口出现。"},{"number":10,"description":"约 20–23s 猎人看到山洞，加速冲刺，猎人—巨兽—洞口同框。"},{"number":11,"description":"约 23–26s 洞内往外拍，猎人扑进洞口翻滚，巨兽冲到洞口。"},{"number":12,"description":"约 26–28s 巨兽撞上洞口岩壁，扬尘糊满画面。"},{"number":13,"description":"约 28–30s 洞内逆光剪影，猎人喘气握弓，巨兽被卡在明亮洞口外。"}],"constraints":"两张设定图严格锁定猎人与 THAROG 外观；箭从命中起一直插在同一只眼；每一秒从上一秒的位置与动作继续，不重置人物、镜头地理和方向；14 秒起猎人始终朝同一处崖壁山洞跑，巨兽沿同一追逐轴线；山洞先在远景出现再成为逃生口；角只在 27–28 秒撞断；避免过度血腥；30.00 秒切黑。缺口：设定图出图词未公开；提示词 image1/image2 编号与发帖顺序相反；成片中箭看起来插在眼睛上方、角断裂细节被扬尘遮住，巨兽角形与设定图不完全一致；音频未转录。","video_prompt":{"title":"猎人 vs THAROG · Seedance 2.5 逐秒提示词","subtitle":"30s · 16:9 · Seedance 2.5 on videoduck · 2 张设定图参考 · 英文完整提示词（作者楼中楼）","content":"$42"}},{"id":"ai-lifehack55-shibuya-billboard-wink-wan3","title":"我怎么上广告牌了 · 涩谷夜景海报眨眼","subtitle":"X · @ai_lifehack55 · ChatGPT 出图 → WAN3.0（SJinn）· 15秒 · 1:1","description":"WAN3.0 夜景广告短片：涩谷路人纷纷回头，女孩抬头发现自己登上巨幅广告牌，海报里的她眨了下眼。","video":"https://media.liuyidaoai.com/tutorials/ai-lifehack55-shibuya-billboard-wink-wan3/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/ai-lifehack55-shibuya-billboard-wink-wan3/poster.jpg","duration":"15秒","durationSec":15,"styleLabel":"时尚广告","shots":7,"references":1,"model":"ChatGPT 图像生成 → WAN3.0（SJinn）","style":"夜晚涩谷街头 · 时尚杂志海报 · 写实电影感广告","aspectRatio":"1/1","sourceUrl":"https://x.com/ai_lifehack55/status/2103685792665329715","sourceAuthor":"@ai_lifehack55","sourcePlatform":"X","sourceImpressions":8250,"sourceStats":{"asOf":"2026-09-27","likes":171,"reposts":20,"bookmarks":63},"formats":["时尚大片"],"hook":{"structure":"路人回头 → 困惑张望 → 抬头看广告牌 → 海报眨眼","opening":"第 0 秒是俯拍的夜晚涩谷十字路口，人潮里一个穿西装的短发女孩正常走着，周围已经有人开始看她。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"约 1.8s 横向跟拍，旁边的西装男回头看她；约 3.5s 正脸特写，她眼神往旁边瞟；约 5s 斜俯拍街道；约 6.3s 斜前方跟拍，她左右张望、一脸困惑。","at":1.8},{"title":"成品揭晓","text":"约 8.3s 更近的脸部特写，视线往上抬；约 9.6s 硬切到大楼外墙的巨幅广告牌，上面是她的「COUTURE」时尚海报，镜头一路推近。","at":8.3},{"title":"结尾怎么收","text":"约 11.5s 推到海报人物的脸，画面静止；约 13.4s 海报里的她单眼眨了一下，随后恢复静止结束。","at":13.4}],"copyThis":"先用一连串短镜头攒「大家为什么都在看我」的疑问，最后抬头把答案放在广告牌上，只让海报里的人眨一次眼收尾。","approx":true},"tags":["15秒 · 时尚广告","1:1 方屏","ChatGPT 出图 → WAN3.0","海报 + 角色设定图 → 成片","广告牌眨眼 · 夜晚涩谷"],"steps":[{"number":1,"title":"第一步：用 ChatGPT 生成时尚杂志海报（image2）","description":"上传一张成人人物照片，把「参考图」卡片 ① 的完整海报指示书贴进 ChatGPT 图像生成，在开头「选择」里选一种 TYPE（1 TREND / 2 SHIFT / 3 MODE / 4 COUTURE）。输出竖版 3:4 的写实时尚杂志海报，只放该 TYPE 固定的一个单词。成片里广告牌上的就是 TYPE 4「COUTURE」。"},{"number":2,"title":"第二步：生成简易角色设定图（image1）","description":"用「参考图」卡片 ② 的角色设定图指示书生成一张 16:9 的简易角色设定图（三视图 + 两张脸部特写），用来锁定主角的脸、发型、体型、服装和鞋。作者未公开设定图，卡片 ② 的配图是成片截帧，不是参考图。"},{"number":3,"title":"第三步：image1 + image2 + 视频提示词交给 WAN3.0","description":"在 SJinn 上选 WAN3.0（或 Seedance2.0），放角色设定图、放海报，贴下方完整视频提示词，提示词里的要改成平台里对应的图片引用。提示词按秒写了 7 个硬切镜头：路人渐渐回头、指她，她越来越困惑，最后抬头看到广告牌，只有海报里的人眨一次眼。作者体感眨眼成功率 WAN3.0 约 40%、Seedance2.0 约 70%，不成功就多抽几次。"}],"references_detail":[{"id":"ai-lifehack55-poster-couture","number":"1","title":"时尚杂志海报 · ① ChatGPT 海报指示书（image2）","subtitle":"作者引用帖示例图 TYPE 4「COUTURE」1086×1448 · 成片广告牌上显示的就是这张","image":"https://media.liuyidaoai.com/tutorials/ai-lifehack55-shibuya-billboard-wink-wan3/refs/ref-poster-couture.jpg","prompt":"$43"},{"id":"ai-lifehack55-character-sheet","number":"2","title":"简易角色设定图 · ② 角色设定图指示书（image1）","subtitle":"作者未公开设定图，此为成片截帧（约 3.8s），不是参考图","image":"https://media.liuyidaoai.com/tutorials/ai-lifehack55-shibuya-billboard-wink-wan3/refs/frame-heroine-NOT-a-reference.jpg","prompt":"$44"}],"storyboard":[{"number":1,"description":"0–1.8s 俯拍夜晚涩谷十字路口，女孩在人潮里正常走，周围有人开始看她。"},{"number":2,"description":"1.8–3.5s 横向跟拍，旁边的西装男回头看她，她还没察觉。"},{"number":3,"description":"3.5–5s 正脸特写，她察觉到视线，眼神往旁边瞟。"},{"number":4,"description":"5–6.3s 斜俯拍街道，路人纷纷看向她。"},{"number":5,"description":"6.3–8.3s 斜前方跟拍，她左右张望，一脸困惑。"},{"number":6,"description":"8.3–9.6s 更近的脸部特写，视线往上抬。"},{"number":7,"description":"9.6–15s 大楼外墙巨幅广告牌显示「COUTURE」海报，镜头推近到海报人物的脸，约 13.4s 她单眼眨眼一次后恢复静止。"}],"constraints":"image1 只继承人物的脸、发型、体型、服装和鞋，设定图里的文字、标签、背景不能出现在画面里；image2 海报要原样显示在广告牌上，字体、背景、配色、构图全部不变；前中段用短镜头硬切，路人反应要克制、写实；结尾只有海报人物眨一次眼，其余保持静止；不要淡入淡出等柔和转场。缺口：作者没公开角色设定图原图；成片里街上的女孩穿黑西装白衬衫，海报人物穿印花吊带裙，衣服对不上；路人指她的动作在成片里不明显；音频没有转录。","video_prompt":{"title":"③ WAN3.0 / Seedance2.0 视频提示词","subtitle":"15s · 1:1 · 角色设定图 + 海报 · 日文完整提示词（主帖）","content":"$45"}},{"id":"studio-oneroom-copypaste-motion-graphics-h3","title":"复制粘贴就能做的动态图形片头 · MiniMax H3","subtitle":"X · @studio_oneroom · MiniMax H3 (2k) · Domo AI · 15秒 · 16:9","description":"MiniMax H3 纯文字生成的扁平动态图形片头：黑白三次反转，字块砸入、碎裂、对撞，15 秒一镜不剪。","video":"https://media.liuyidaoai.com/tutorials/studio-oneroom-copypaste-motion-graphics-h3/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/studio-oneroom-copypaste-motion-graphics-h3/poster.jpg","duration":"15秒","durationSec":15,"styleLabel":"扁平动态图形","shots":9,"references":0,"model":"MiniMax H3 (2k)（Domo AI）","style":"扁平图形动态设计 · 白 / 黑 / 橙三色 · 粗几何无衬线字","aspectRatio":"16/9","sourceUrl":"https://x.com/studio_oneroom/status/2092220604309205341","sourceAuthor":"@studio_oneroom","sourcePlatform":"X","sourceImpressions":167166,"sourceStats":{"asOf":"2026-09-27","likes":988,"reposts":103,"bookmarks":1441},"formats":["字效·片头"],"hook":{"structure":"线条漩涡 → 标题拼装 → 字块连砸 → 品牌对撞定版","opening":"第 0 秒黑底上白色线条从四周涌向中心形成漩涡，约 0.5s 一个巨大的橙色圆砸进画面。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"约 1–2s 线条拼成「MiniMax H3」，底色翻成白；约 2.5s 急推到橙色「H3」占满全屏；约 4.5s 竖条把画面切成三列；约 5.5s「MOTION」整屏砸入。","at":1},{"title":"节奏加速","text":"约 6.5s 硬切黑底，白色线框网格往纵深延伸，「GRAPHICS」沿网格成片复制；约 8–9.5s 字卡连闪「NO EDIT」「ONE SHOT」「コピペで」「超簡単！」。","at":6.5},{"title":"结尾怎么收","text":"约 10.5s 橙底上「ONEROOM」从两侧对撞成一行，碎片像玻璃一样四散；约 12.5s 翻回白底，「AI STUDIO」在上、「ONEROOM」在下，中间一条橙色短线，静止到结束。","at":10.5}],"copyThis":"开头把 9 段要出现的文字全列成变量，只锁 3 色、1 种字体和 3 层不同速度，每镜只写一句动作加时间点，细节交给模型发挥。","approx":true},"tags":["15秒 · 片头 / 标题 Logo","16:9 横屏","MiniMax H3 (2k)","纯文生视频 · 无参考图","改开头 9 个字符串就能套用"],"steps":[{"number":1,"title":"第一步：改掉 subject_definitions 里的 9 个字符串","description":"下方完整提示词开头定义了强调色（饱和橙）和第 1–9 个字符串（MiniMax H3 / MOTION / GRAPHICS / NO EDIT / ONE SHOT / コピペで / 超簡単！ / ONEROOM / AI STUDIO），换成你自己的标题、口号和品牌名即可；分屏竖条取自第 1 个字符串里大写 H 的竖笔。"},{"number":2,"title":"第二步：整段贴进 MiniMax H3，纯文字生成","description":"不需要参考图。作者在 Domo AI 上用 MiniMax H3 (2k) 生成，说生成偏慢、但画面很干净。提示词锁死了只用 9 个字符串、1 种粗几何无衬线字、白 / 黑 / 橙 3 色、底色黑白反转 3 次，以及背景、装饰、文字 3 层用不同速度运动。"},{"number":3,"title":"第三步：少写细节，多抽几次","description":"作者的经验是这类提示词不要写太满，留给 H3 自由发挥再多抽几次更有意思，这次特意删到不足 3000 字。每镜只给时间点和一句动作（00:01 拼标题 … 00:12 定版），最后 13.8–15s 全画面静止。"}],"references_detail":[],"storyboard":[{"number":1,"description":"0–1s 黑底，白色线条从四边涌向中心形成漩涡，巨大橙色圆砸入，画面震动。"},{"number":2,"description":"1–2.5s 线条拼成「MiniMax H3」，底色由黑翻白，「H3」为橙色。"},{"number":3,"description":"2.5–3.5s 急推到橙色「H3」，放大到占满全屏。"},{"number":4,"description":"3.5–5.2s 白底，竖条贯穿全高，把画面切成三列上下滚动，其余字母被甩出画面。"},{"number":5,"description":"5.2–6.4s 三列急停，「MOTION」整屏砸入并定住。"},{"number":6,"description":"6.4–8s 硬切黑底，白色线框网格倾斜延伸向纵深，「GRAPHICS」沿网格成片复制涌来。"},{"number":7,"description":"8–10s 字卡快闪：「NO EDIT」「ONE SHOT」「コピペで」「超簡単！」。"},{"number":8,"description":"10–12s 橙底，「ONEROOM」从左右两侧高速对撞成一行，碎片像玻璃一样四散。"},{"number":9,"description":"12–15s 黑底「ONEROOM」落到下半屏，随后翻白底，「AI STUDIO」在上、中间一条橙色短线，静止到结束。"}],"constraints":"画面上只能出现 9 个指定字符串，拼写、大小写、空格都要一字不差，不能乱码、多字或少字；「第 1 个字符串」这类称呼本身不能画到画面上；只用 1 种粗几何无衬线字体（日文用粗黑体），只用米白、墨黑、强调色 3 色；底色黑白反转 3 次；背景、装饰、文字 3 层始终以不同速度运动，只有最后 13.8–15s 全部静止；背景是空心描边的「ONEROOM」超低速从右往左流动；9 镜构图不能重复；字形不能随机故障。与成片不符：提示词只让第 1 个字符串的最后一个字是强调色，成片里「H3」两个字都是橙色；Shot 7 要求一卡一词，成片里「コピペで」和「ONE SHOT」、「超簡単！」出现在同一张卡上。","video_prompt":{"title":"MiniMax H3 完整提示词（纯文生视频）","subtitle":"15s · 16:9 · 无参考图 · 日文完整提示词（主帖长文）","content":"$46"}},{"id":"ai-lifehack55-miniature-coronation-portrait-wan3","title":"微缩城市的戴冠式 · 巨幅公主肖像揭幕","subtitle":"X · @ai_lifehack55 · WAN3.0（SJinn）· 15秒 · 1:1","description":"WAN3.0 微缩城市延时：广场上的小人搭起脚手架、挂上巨幕，烟花升空后幕布落下，露出公主巨幅肖像。","video":"https://media.liuyidaoai.com/tutorials/ai-lifehack55-miniature-coronation-portrait-wan3/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/ai-lifehack55-miniature-coronation-portrait-wan3/poster.jpg","duration":"15秒","durationSec":15,"styleLabel":"微缩模型","shots":11,"references":0,"model":"WAN3.0（SJinn）","style":"微缩城市大场景 · 施工进度跳切 · 烟花揭幕","aspectRatio":"1/1","sourceUrl":"https://x.com/ai_lifehack55/status/2103806816731681212","sourceAuthor":"@ai_lifehack55","sourcePlatform":"X","sourceImpressions":2240,"sourceStats":{"asOf":"2026-09-27","likes":73,"reposts":9,"bookmarks":7},"formats":["拆装·制作过程"],"hook":{"structure":"空广场 → 搭架装框 → 挂幕布 → 烟花 → 幕落揭幕","opening":"第 0 秒高位斜俯拍一座密密麻麻的微缩城市，中央广场只堆着建材，小人们正把木料往里搬。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"每一刀都跳一大段工期：约 1.5s 多层脚手架已搭好、工人在顶层固定横杆；约 2.5s 巨大的竖版金色画框吊装到位；约 4s 工人把整块幕布拉开盖住画框；约 6s 广场挂满彩旗，人群开始聚集。","at":1.5},{"title":"蓄势","text":"约 7.5s 镜头升高后拉，广场、街区和远处城市全部入画，满场观众；约 9.5s 天色转暗，一发烟花从城市深处升空，约 10.5s 在画框上空炸开。","at":7.5},{"title":"成品揭晓","text":"约 11.2s 幕布往下滑落，露出头戴王冠、身穿白色礼服的公主肖像；约 12.3s 切近到整幅画框，观众举手欢呼，纸屑飘落到结束。","at":11.2}],"copyThis":"每个镜头只拍当前工期剩下的最后一步，一刀跳一大段进度；肖像一直用幕布盖住，攒到烟花炸开后再一次性揭开。","approx":true},"tags":["15秒 · 微缩城市","1:1 方屏","WAN3.0（SJinn）","1 张 3:4 图 → 揭幕肖像","施工跳切 · 烟花揭幕"],"steps":[{"number":1,"title":"第一步：需自备一张 3:4 竖图作 image1","description":"需自备一张 3:4 竖图作 image1，它只作为最后揭幕的巨幅肖像内容，不影响城市、建筑、居民和装饰。作者没有公开自己用的这张图，也没有给出图提示词；成片里揭开的是一位头戴王冠、身穿白色礼服、手持权杖的公主全身像。"},{"number":2,"title":"第二步：image1 + 下方完整提示词交给 WAN3.0","description":"在 SJinn 上选 WAN3.0，把提示词里的改成平台里对应的图片引用，贴下方完整提示词。它按 11 个镜头写死了施工进度（0% → 30% → 50% → 70% → 85% → 100%），每个镜头从已完成的部分开始，只拍剩下的最后一步；全程用远景大广角，肖像在揭幕前一直被不透明幕布完全遮住。"},{"number":3,"title":"第三步：烟花、揭幕和收尾按秒控制","description":"9.5s 起一发烟花升空、10.7s 炸开，11.5s 幕布整块落下，12.3–15s 镜头缓慢推近，但始终让 3:4 画框完整留在画面里。作者提醒生成结果会有波动、复现性基本没验证过，WAN3.0 也还做不好高速蒙太奇。"}],"references_detail":[],"storyboard":[{"number":1,"description":"0–1.5s 高位斜俯拍微缩城市，中央广场只有建材，工人把木料、纸板、绳子运进来。"},{"number":2,"description":"1.5–2.5s 跳切：多层脚手架已搭好，工人在顶层固定最后一根横杆。"},{"number":3,"description":"2.5–3.8s 跳切：脚手架里吊起巨大的竖版金色画框，工人扶着固定顶梁。"},{"number":4,"description":"3.8–5s 跳切：工人把幕布从上往下展开，盖满画框正面。"},{"number":5,"description":"5–6s 侧面高位：外侧脚手架大多已拆，工人挂最后的彩旗和纸饰。"},{"number":6,"description":"6–7.3s 远景：彩旗挂满，工人离场，居民开始聚集到广场。"},{"number":7,"description":"7.3–9.5s 最大远景，镜头升高后拉：广场挤满观众，街区和远处城市全部入画。"},{"number":8,"description":"9.5–10.5s 天色转暗，一发烟花从城市深处升空，观众抬头看。"},{"number":9,"description":"10.5–11.2s 烟花在画框上空炸开，幕布仍然盖着。"},{"number":10,"description":"11.2–12.3s 幕布往下滑落，露出头戴王冠、身穿白色礼服的公主肖像。"},{"number":11,"description":"12.3–15s 切近到整幅画框，观众举手欢呼，纸屑飘落，定格到结束。"}],"constraints":"image1 只用作最后揭幕的肖像内容，揭幕前不能露出，幕布不能透光，也不能把参考图的内容用到城市、建筑、居民和装饰上；城市要做出剪纸、纸板、布料拼贴的手工质感，不要光滑的 3DCG 模型或游戏渲染；全程远景大广角，不能只拍画框；不要魔法式的自动建造，不要画面文字、Logo、字幕；只能出现一幅肖像。与成片不符：成片更像普通微缩模型，纸张拼贴的质感不明显；约 12.3s 是直接切近到画框，不是提示词写的从远景缓慢推近，最后画框约占画面高度 95%，超过要求的 75–80%；烟花段天色突然变暗，揭幕后又回到白天；前景道路上不止一辆车。缺口：作者没公开 image1 原图，也没有出图提示词；音频没有转录。","video_prompt":{"title":"WAN3.0 完整视频提示词","subtitle":"15s · 1:1 · 3:4 竖版肖像图 · 日文完整提示词（主帖长文）","content":"$47"}},{"id":"magnific-mayday-hot-sauce-claude-opus","title":"一滴辣酱，全员警报 · MAYDAY 辣酱广告","subtitle":"X · @magnific · Claude Opus 5.5 → Magnific · 30秒 · 16:9","description":"Claude Opus 5.5 写分镜、Magnific 出片的辣酱广告：一滴辣酱落向塔可，控制室拉响最高警报。","video":"https://media.liuyidaoai.com/tutorials/magnific-mayday-hot-sauce-claude-opus/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/magnific-mayday-hot-sauce-claude-opus/poster.jpg","duration":"30秒","durationSec":30,"styleLabel":"写实广告","shots":6,"references":4,"model":"Claude Opus 5.5（写提示词）→ Magnific","style":"复古 70 年代控制室 · 钴蓝 / 信号红 / 奶油色 · 海报大字叠实拍","aspectRatio":"16/9","sourceUrl":"https://x.com/magnific/status/2103889116022931629","sourceAuthor":"@magnific","sourcePlatform":"X","sourceImpressions":10798,"sourceStats":{"asOf":"2026-09-27","likes":108,"reposts":11,"bookmarks":80},"formats":["产品广告"],"hook":{"structure":"辣酱将滴 → 控制室警报 → 全力抢险 → 咬一口破防 → 任务完成 → 产品定版","opening":"第 0 秒是极近的慢镜头：红色辣酱瓶斜在一只脆皮塔可上方，瓶口一滴辣酱正在成形，约 3s 巨大的红字「T-MINUS 3」铺满背景。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"约 5s 切到复古控制室，留小胡子的主管起身拍下红色按钮，约 7s 满屏「SPICE LEVEL / CRITICAL」；约 10s 巨型风扇全速转动，约 12s 闸门放水，约 13s 红字「EMERGENCY」横扫画面。","at":5},{"title":"包袱","text":"约 15s 那一滴终于落到塔可上；约 17s 卷发小哥咬一口、若无其事地嚼，约 19s 突然瞪大眼睛，叠上「too MUCH」。","at":15},{"title":"结尾怎么收","text":"约 20s 控制室欢呼、纸片乱飞，约 23s 小哥含泪竖大拇指，「MISSION ACCOMPLISHED」配手写「barely」；约 25s 钴蓝底上单瓶定版，身后是巨大品牌字，下方「ONE DROP. FULL ALERT.」。","at":20}],"copyThis":"把「吃辣」这件小事当成灾难片来拍：每镜都配一句铺满全屏的大字，四角加小编号标签，全程只用 3 种颜色，最后回到一瓶酱的定版。","approx":true},"tags":["30秒 · 产品广告","16:9 横屏","Claude Opus 5.5 → Magnific","简版创意 + 4 张风格参考 → 6 镜提示词","辣酱 · 控制室警报喜剧"],"steps":[{"number":1,"title":"第一步：一句创意 + 4 张风格参考图交给大模型","description":"Magnific 给 Claude Opus 5.5 和 GPT-6 Astra 的是同一句简版创意：30 秒辣酱广告，一滴辣酱落到塔可上，引发控制室紧急事态，动作片式幽默、原创品牌和动态图形，英文字幕、无对白，参考但不照抄。另附 4 张风格参考图，它们的出图提示词写在帖子每张图的 ALT 文字里，见「参考图」卡片。"},{"number":2,"title":"第二步：让模型写出 6 镜分镜提示词","description":"Claude 把品牌定为 MAYDAY，写成 6 个 5 秒镜头（T-MINUS 3 → SPICE LEVEL CRITICAL → EMERGENCY → TOO MUCH → MISSION ACCOMPLISHED → ONE DROP. FULL ALERT.）。每镜都重复同一段风格描述（电影机、光晕、胶片颗粒、钴蓝 / 信号红 / 奶油色）、四角小标签文字、一句满屏大字和声音，并写明无对白、无黑边。"},{"number":3,"title":"第三步：在 Magnific 里逐镜生成并拼成 30 秒","description":"把下方完整提示词按镜头在 Magnific 里生成。帖子没说 Magnific 内部具体用了哪个视频模型。成片左上角有「Claude Opus 5.5」标签、右上角有 Magnific 水印；同线程还有 GPT-6 Astra 版（品牌 GOTA），可以对比同一个创意的不同写法。"}],"references_detail":[{"id":"magnific-ref-01-stands-nugget","number":"1","title":"看台观众与蘸酱的她 · 出图提示词（图片 ALT 文字）","subtitle":"作者原图 2688×1520 · 编辑风看台群像","image":"https://media.liuyidaoai.com/tutorials/magnific-mayday-hot-sauce-claude-opus/refs/ref-01-stands-nugget.jpg","prompt":"Wide 16:9 editorial campaign photo: ten stylish spectators packed in three tiers of outdoor stands, all looking right at an off-frame match. Hazy pale blue seaside sky, warm golden-hour sun. Preppy vintage tennis-club wardrobe in cream, butter yellow, forest green, navy and sky blue: cable-knit cricket sweaters and vests with striped V-necks, polo collars, green Harrington jacket, yellow windbreaker, sweater tied over shoulders, white bucket hat, white cap, navy cap. Almost all wear retro tortoiseshell or black sunglasses. Diverse ages and ethnicities, including a mustached man, a woman with a dark bob and gold hoops, a bearded older man with salt-and-pepper hair, a blonde with a scrunchie ponytail, a grey-haired woman with a low bun. Everyone serious except a curly red-haired woman in a bright turquoise terry zip jacket, eyes closed, blissfully dipping a chicken nugget into a ketchup cup. Medium format film, soft grain, faded warm grade."},{"id":"magnific-ref-02-office-football","number":"2","title":"办公室走廊踢球 · 出图提示词（图片 ALT 文字）","subtitle":"作者原图 2688×1520 · 35mm 低机位广角 · 注意：这段 ALT 正好 1000 字符（X 上限），「holding a roll」后面接的是另一段描述，原文不完整","image":"https://media.liuyidaoai.com/tutorials/magnific-mayday-hot-sauce-claude-opus/refs/ref-02-office-football.jpg","prompt":"Wide 16:9 editorial fashion photo on 35mm film, low wide-angle from floor level, slight Dutch tilt. Late-90s office hallway: beige walls, drop ceiling, grey filing cabinets, burnt orange carpet. Left foreground, a man with long dark wavy hair kicks a football mid-strike, leaning back, leg toward camera; royal blue sleeveless jersey with white trim, red shorts with white stripes, white crew socks, chunky white retro sneakers. The ball flies, slightly blurred. Back right, an open steel elevator is the goal: a dark-haired goalkeeper crouches inside in ready stance, white gloves, white retro football jersey, blue shorts, sneakers, eyes on the ball. Middle background, a couple leans on the wall chatting, ignoring it: a man in a mint-green oversized V-neck sweater and grey trousers, a blonde woman with messy updo in black turtleneck, black pleated mini skirt and heels holding a roll adults in an office space, one person kicks a soccer ball while another prepares to catch it near an elevator."},{"id":"magnific-ref-03-poster-todo-cambia","number":"3","title":"「TODO CAMBIA」字效海报 · 出图提示词（图片 ALT 文字）","subtitle":"作者原图 2048×1152 · 橙色压字 + 草坡照片","image":"https://media.liuyidaoai.com/tutorials/magnific-mayday-hot-sauce-claude-opus/refs/ref-03-poster-todo-cambia.jpg","prompt":"Wide 16:9 editorial poster. Background: 35mm film photo of a vast, gently curved green grassy hill under a deep blue sky with wispy cirrus clouds. A tiny lone man stands on the crest at the right third, looking down: red knit sweater, off-white trousers, navy cap, hands in pockets, dwarfed by empty space. Over the photo, a massive condensed all-caps grotesque sans-serif headline fills the canvas edge to edge in three tightly stacked lines: \"TODO CAMBIA\" / \"CUANDO\" / \"LO MIRAS BIEN\". Letters in warm saturated orange with crumpled paper risograph texture, white speckles, scuffed ink dropouts and slight transparency so sky and grass show through. A loose hand-drawn cream marker loop circles \"TODO\" and the \"C\" of \"CUANDO\". Tiny centered white spaced-out caps at the bottom: \"OTRA PERSPECTIVA\". Crinkled paper overlay, film grain, muted analog colors, faded 90s print feel. Bold graphic design meets documentary photography."},{"id":"magnific-ref-04-poster-otra-mirada","number":"4","title":"「OTRA MIRADA」字效海报 · 出图提示词（图片 ALT 文字）","subtitle":"作者原图 2048×1152 · 动态模糊奔跑 + 奶油色粗圆体","image":"https://media.liuyidaoai.com/tutorials/magnific-mayday-hot-sauce-claude-opus/refs/ref-04-poster-otra-mirada.jpg","prompt":"Wide 16:9 cinematic poster. Background: heavily motion-blurred photo of a woman running in profile left to right along a beach at golden hour, long-exposure panning shot, her silhouette smeared into horizontal streaks of warm orange and dark brown with ghosted trails. Deep teal-green sky in the upper two thirds, a pale blurred light band across the middle, burnt orange sand below, dark vignette. Heavy analog grain, saturated 70s color grade. Typography: ultra-bold wide rounded heavy sans-serif in cream, all caps, tight tracking. Upper left: \"OTRA\" / \"MIRADA\" stacked. Lower right: \"MISMO\" / \"MUNDO\" stacked, overlapping the runner's legs. Small spaced-out cream caps top right: \"CAMBIA LA PERSPECTIVA\". Small spaced-out cream caps bottom left: \"TODO EMPIEZA EN CÓMO LO VES.\" Sharp clean type against the dreamy blurred photo, editorial sports-campaign aesthetic."}],"storyboard":[{"number":1,"description":"0–5s 极近慢镜头：红色辣酱瓶斜在塔可上方，一滴辣酱成形、颤动、滴落，背景铺满红字「T-MINUS 3」。"},{"number":2,"description":"5–10s 复古 70 年代控制室，红色连体服操作员，留小胡子的主管拍下红色按钮，红色警灯转起，满屏「SPICE LEVEL / CRITICAL」。"},{"number":3,"description":"10–15s 动作蒙太奇：巨型风扇全速转动、钢闸门放出水墙、操作员双手拉杆，红字「EMERGENCY」横扫，叠红色手写字。"},{"number":4,"description":"15–20s 辣酱落在塔可上，卷发红运动服小哥咬一口，先淡定后瞪眼、流汗、头发被风吹起，急推脸部，叠「too MUCH」。"},{"number":5,"description":"20–25s 快切：控制室欢呼拥抱、纸片乱飞，主管擦汗，小哥含泪竖大拇指，「MISSION ACCOMPLISHED」配手写「barely」。"},{"number":6,"description":"25–30s 钴蓝底单瓶产品定版，身后巨大红色品牌字，下方奶油色「ONE DROP. FULL ALERT.」。"}],"constraints":"每镜都重复同一段风格描述：电影机拍摄、光晕、暖调胶片颗粒、钴蓝 / 信号红 / 奶油色、海报式大字叠实拍、四角小标签；画面里只能有一瓶酱，不出现其他品牌；无对白，只有音效和音乐；无黑边、无强暗角；第 5 镜用全画幅，不分屏、不用圆形画框。与成片不符：定版的品牌字和瓶身标签是「MAYAM」，不是提示词写的「MAYDAY」；「EMERGENCY」上方的手写字不是提示词写的「obviously」，拼写不清；第 5 镜四角标签的编号有错（左下写成 SP-03）。缺口：帖子没说 Magnific 内部用的是哪个视频模型；4 张参考图是给大模型的风格参考，帖子没说是否直接喂给了视频模型；参考图 2 的 ALT 正好 1000 字符，原文不完整；音频没有转录。","method_prompt":{"title":"给大模型的创意简报（Creative brief）","subtitle":"Claude Opus 5.5 与 GPT-6 Astra 拿到的同一句创意 · 英文原文 · 帖子称为 short version","content":"Create a cinematic 30-second hot sauce ad. One drop hits a taco and triggers a control-room emergency. Action-movie humor, original branding and motion graphics. English text, no dialogue. Take inspiration from the references without copying them.","note":"来源：@magnific 同一线程第二条帖（https://x.com/magnific/status/2103889158972580030），帖子原话：“We gave Claude and GPT an idea and a few visual references. Here’s the short version so you don’t have to start from scratch:” 引号里就是这段 brief，帖子称之为 short version（精简版）。\n用法：Magnific 把这同一句创意，连同 4 张风格参考图（出图提示词在帖子每张图的 ALT 里，见下方「参考图」卡片），分别交给 Claude Opus 5.5 和 GPT-6 Astra，让模型自己发展方向、写出逐镜提示词；本页「步骤三」就是 Claude Opus 5.5 写出来的那一版。帖子结尾原话：“Each model developed its own direction, watch both below / Prompts in ALT”。"},"video_prompt":{"title":"Claude Opus 5.5 写的 6 镜视频提示词","subtitle":"30s · 16:9 · 6 × 5s · 英文完整提示词（线程第 3 帖长文）","content":"$48"}},{"id":"iamsofiaijaz-museum-steps-frozen-crowd-seedance","title":"博物馆台阶上的静止人群 · Seedance 2.5","subtitle":"X · @iamsofiaijaz · Seedance 2.5 · OpenArt · 30秒 · 16:9","description":"Seedance 2.5 两镜超现实短片：女主穿过台阶上静止的古装人群站到正中，下一镜全场疯狂甩身、她纹丝不动。","video":"https://media.liuyidaoai.com/tutorials/iamsofiaijaz-museum-steps-frozen-crowd-seedance/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/iamsofiaijaz-museum-steps-frozen-crowd-seedance/poster.jpg","duration":"30秒","durationSec":30,"styleLabel":"超现实","shots":2,"references":0,"model":"Seedance 2.5（OpenArt）","style":"写实古装群演 · 静止人群与失控人群 · 冷调胶片感","aspectRatio":"16/9","sourceUrl":"https://x.com/iamsofiaijaz/status/2103685419695280351","sourceAuthor":"@iamsofiaijaz","sourcePlatform":"X","sourceImpressions":1251,"sourceStats":{"asOf":"2026-09-26","likes":22,"reposts":0,"bookmarks":10},"formats":["角色表演","电影叙事"],"hook":{"structure":"近景抽烟 → 拉远穿过静止人群 → 全场失控而她不动","opening":"第 0 秒是戴细框眼镜的女子正脸大特写，叼着烟吸一口，约 2–3s 把烟雾直接吐向镜头，糊住整个画面。","openingAt":0,"beats":[{"title":"静止人群揭晓","text":"约 3–6s 烟散开、镜头后拉，身后台阶上挤满红军装、芭蕾舞裙、胸甲的古装人群，全部一动不动盯着镜头。","at":3},{"title":"过程怎么推进","text":"约 6–13s 她转身背对镜头，从人群缝里侧身挤上台阶，没人让路，走到正中后回身站定。","at":6},{"title":"几段怎么切换","text":"约 14s 硬切到远景：全场人群突然前后弯腰甩身、此起彼伏，镜头一路拉远到整座博物馆立面，只有正中的她站着不动。","at":14}],"copyThis":"第一镜让所有人绝对静止、只有主角在动；第二镜反过来，所有人失控、只有主角不动——同一套站位，动静对调。","approx":true},"tags":["30秒 · 两镜","16:9 横屏","Seedance 2.5","OpenArt","静止人群","动静对调"],"steps":[{"number":1,"title":"准备主角参考图，写死所有群演站位","description":"两段提示词都以「参考图里的角色就是主角，脸、发型、服装每一帧保持一致」开头（作者没有公开这张参考图，跟做需自备）。然后按前排、第二排、第三排从左到右逐个写群演装扮，并把第三排正中标成主角的站位；第二镜要求「完全同一批人、同一位置」。"},{"number":2,"title":"SHOT 1（14 秒）：全场静止，只有主角移动","description":"时间轴：0–3s 正脸特写抽烟吐烟；3–6s 烟散镜头后拉，露出身后静止的人群；6–7s 转身；7–12s 以正常步速侧身挤过人群上台阶（写明路线经过哪些人）；12–14s 到达正中、回身站定。反向约束：人群绝不移动、不让路、不留通道，主角不能慢动作。"},{"number":3,"title":"SHOT 2（15 秒）：全场失控，只有主角不动","description":"第一帧就是远景且人群已在剧烈甩身：头后仰到底 → 上身猛折到膝盖 → 再弹回，循环不停，不同区块错开相位形成波浪；0–10s 镜头后拉上升到整座立面，10–15s 锁定。主角全程站在正中不动。两镜都要求无音乐，只留环境声。两段分别生成后在第一次动作处硬切拼接。"}],"references_detail":[],"storyboard":[{"number":1,"description":"SHOT 1（约 0–14s）：正脸特写吸烟吐烟 → 后拉露出台阶上静止的古装人群 → 转身挤过人群上台阶 → 在第三排正中回身站定。"},{"number":2,"description":"SHOT 2（约 14–30s）：硬切远景，全场人群前后弯腰甩身、此起彼伏，镜头拉远上升到整座博物馆立面后锁定，主角在正中始终不动。"}],"constraints":"主角脸、发型、服装按参考图全程一致；群演站位、服装、顺序两镜完全相同；第一镜人群绝对静止、不让路、不留通道，主角正常步速不慢动作；第二镜人群全程剧烈甩身不停、不整齐划一，主角不动不跳舞；写实真人质感，禁止油画/插画/CGI/蜡像感；无音乐、无字幕水印。缺口：作者未公开主角参考图；成片人群以重复的红军装、芭蕾裙、胸甲为主，提示词里的名画人物基本认不出；音频未转录。","video_prompt":{"title":"静止人群两镜 · Seedance 2.5 提示词","subtitle":"SHOT 1 14s + SHOT 2 15s · 16:9 · Seedance 2.5 on OpenArt · 需主角参考图 · 英文完整提示词（作者楼中楼）","content":"$49"}},{"id":"aimikoda-sky-duel-seedance-2-5","title":"红发双刀 vs 单刀武士 · 云端空战剑斗 · Seedance 2.5","subtitle":"X · @aimikoda · Seedance 2.5 · 30秒 · 16:9","description":"Seedance 2.5 二次元空战剑斗：双刀红发剑士与单刀武士六镜追打，逆光剪影与彩色近身缠斗交替切换。","video":"https://media.liuyidaoai.com/tutorials/aimikoda-sky-duel-seedance-2-5/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/aimikoda-sky-duel-seedance-2-5/poster.jpg","duration":"30秒","durationSec":30,"styleLabel":"手绘动漫","shots":6,"references":2,"model":"Seedance 2.5","style":"手绘厚涂二次元 · 高空追逐剑斗 · sakuga 作画","aspectRatio":"16/9","sourceUrl":"https://x.com/aimikoda/status/2103512083778216062","sourceAuthor":"@aimikoda","sourcePlatform":"X","sourceImpressions":2508,"sourceStats":{"asOf":"2026-09-26","likes":51,"reposts":0,"bookmarks":27},"formats":["电影叙事","角色表演"],"hook":{"structure":"剪影追击 → 彩色近身缠斗 → 剪影回收拉远","opening":"第 0 秒就是淡蓝天空里两道逆光黑剪影对冲：长发双刀的一方从左上俯冲，单刀武士从右下迎上，刀已经碰在一起——没有铺垫直接开打。","openingAt":0,"beats":[{"title":"剪影追击怎么推进","text":"约 0–10s 全程逆光剪影，镜头横向跟拍两人在空中追砍、翻滚、互相绕到身后，只靠轮廓和刀线读动作。","at":3},{"title":"转彩色近身缠斗","text":"约 10.5s 突然切进彩色近景：红发 Erza 双刀与棕衣 Saito 贴身交错，约 15s 刀刃相撞迸出火花，之后连续翻身、下坠、换位。","at":10.5},{"title":"结尾怎么收","text":"约 26s 回到剪影对冲，约 27.5s 猛然拉远到云海全景：两条细长的刀痕弧线划过天空，两人变成小黑点继续飞开。","at":27.5}],"copyThis":"六镜都写清「谁从哪条路线逃、谁封住哪个出口」，再用一个开场同构的大全景收尾：两道交叉刀痕把天空「划开」。","approx":true},"tags":["30秒 · 六镜空战","16:9 横屏","Seedance 2.5","双角色参考图","Midjourney 角色","sakuga 剑斗"],"steps":[{"number":1,"title":"先准备两张角色参考图，只借外观和画风","description":"作者用自己引用帖里 Midjourney v8.2 出的两张角色图：@[char1 ref] = Erza（红发、双刀），@[char2 ref] = Saito（单刀武士）。提示词第一段就声明：参考图只提供外观、武器和厚涂动漫画风，不提供姿势——要求重新画全身动作轮廓，不要把参考图里的蹲姿平移或旋转。"},{"number":2,"title":"设置 30 秒 · 六镜 · 16:9，把一句话剧情写在最前","description":"开头先交代总剧情：Saito 反复从 Erza 左臂下方钻到身后偷袭；被他踢翻后，Erza 故意露出同一个空档当诱饵，用第二把刀封住出口。再写统一的动作规范（实时 sakuga、不慢动作、不长时间对刀）和「刚好三把刀」的武器约束，防止刀凭空多出或脱手乱飞。"},{"number":3,"title":"逐镜写时间段 + 路线，最后写画风与声音","description":"Shot 1–6 按 0-4s / 4-10s / 10-14s / 14-20s / 20-25s / 25-30s 写，每镜都说明镜头怎么跟（斜向追、横向跟拍、侧面全身、跟着下坠等）和两人路线变化；Shot 6 复刻开场斜向构图并拉远，交叉刀痕划破云层收尾。末段写淡蓝灰天空、手绘阴影、无能量光束，音效只要破空、布料、金属声，无对白、字幕、音乐。"}],"references_detail":[{"id":"ref-aimikoda-erza","number":"1","title":"@[char1 ref] · Erza（红发双刀）","subtitle":"引用帖 Midjourney v8.2 · 第 3 张","image":"https://media.liuyidaoai.com/tutorials/aimikoda-sky-duel-seedance-2-5/refs/ref-erza.jpg","prompt":"作者引用帖原文（只公开了风格参数，没有公开这张图的完整出图词）：\nMidjourney v8.2\n\nA new mix:\n--sref 1448908625 3123598145 387469134 3207844525 --profile kxxcnp9 --stylize 250\n\nAnd a tip for poses: Use keywords like mid-air action pose, mid-air combat pose for more dynamic poses.\n\n注：作者未点名 4 张图中哪两张是 char1/char2；按提示词描述（Erza 双刀）与成片比对判定为第 3 张。"},{"id":"ref-aimikoda-saito","number":"2","title":"@[char2 ref] · Saito（单刀武士）","subtitle":"引用帖 Midjourney v8.2 · 第 2 张","image":"https://media.liuyidaoai.com/tutorials/aimikoda-sky-duel-seedance-2-5/refs/ref-saito.jpg","prompt":"作者引用帖原文（只公开了风格参数，没有公开这张图的完整出图词）：\nMidjourney v8.2\n\nA new mix:\n--sref 1448908625 3123598145 387469134 3207844525 --profile kxxcnp9 --stylize 250\n\nAnd a tip for poses: Use keywords like mid-air action pose, mid-air combat pose for more dynamic poses.\n\n注：作者未点名 4 张图中哪两张是 char1/char2；按提示词描述（Saito 单刀）与成片比对判定为第 2 张。"}],"storyboard":[{"number":1,"description":"0–4s 斜向追击大全景：Erza 从左上俯冲下劈，Saito 从右下迎上格挡，从她左臂下方滚到身后（成片此段为逆光剪影）。"},{"number":2,"description":"4–10s 横向跟拍追砍：Erza 连续转胯挥刀，Saito 后仰闪避后反撩，再次钻到她左肩后方（仍为剪影）。"},{"number":3,"description":"10–14s 侧面全身（成片约 10.5s 切进彩色）：Saito 从身后划破 Erza 左袖并侧踢她上臂，Erza 翻滚下坠、回头盯住对手收刀调整。"},{"number":4,"description":"14–20s 跟随下坠：Erza 重复右手下劈、故意露出左臂下空档，Saito 走老路线时她用蓄着的左刀封住出口，刀刃相撞迸出火花（约 15s），他被迫螺旋翻身闪开。"},{"number":5,"description":"20–25s 上升：Saito 改为头顶下劈，Erza 侧身滚到刀下、右刀引开、左刀横扫回程路线，擦碰后两人沿相反弧线分开又再次对冲。"},{"number":6,"description":"25–30s 复刻开场斜向构图（成片回到剪影）：交叉一击后猛拉远到云海全景，两道细长刀痕交叉划过天空，两人仍在飞行中。"}],"constraints":"参考图只提供外观、武器与厚涂画风，不提供姿势；全程刚好三把刀（Erza 双手各一、Saito 右手一把、左手空着），刀柄不离手，禁止掉落/合并/复制/漂浮武器；实时 sakuga，禁慢动作、顿帧、对视僵持、长时间对刀、重复蹲姿轮廓；两人始终在空中移动，接触与踢击改变轨迹；淡蓝灰天空、漫射日光、手绘阴影，无能量光束或传送门；只要破空/布料/金属/踢击音效，无对白、字幕、屏幕文字或音乐。缺口：引用帖只公开 Midjourney 风格参数，未公开两张角色图出图词；char1/char2 对应哪张图为比对判断；成片开头与结尾的逆光剪影段提示词未要求，是模型自发处理。","video_prompt":{"title":"Erza vs Saito 云端空战 · Seedance 2.5 提示词","subtitle":"30s · 16:9 · Seedance 2.5 · 2 张角色参考图 · 六镜英文完整提示词（作者楼中楼）","content":"$4a"}},{"id":"just-sharon7-samurai-cat-reeds-seedance-2-5","title":"斗笠武士橘猫 · 芦苇荡斗忍者 · Seedance 2.5","subtitle":"X · @Just_sharon7 · Seedance 2.5 · TapNow · 30秒 · 16:9","description":"Seedance 2.5 写实武士猫动作短片：斗笠橘猫在芦苇荡连闪忍者、空中翻腾，落上小船后直立走向镜头。","video":"https://media.liuyidaoai.com/tutorials/just-sharon7-samurai-cat-reeds-seedance-2-5/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/just-sharon7-samurai-cat-reeds-seedance-2-5/poster.jpg","duration":"30秒","durationSec":30,"styleLabel":"写实动作","shots":8,"references":0,"model":"Seedance 2.5（TapNow）","style":"写实毛发 · 日式芦苇荡武侠动作 · 史诗又荒诞","aspectRatio":"16/9","sourceUrl":"https://x.com/Just_sharon7/status/2103453974980591826","sourceAuthor":"@Just_sharon7","sourcePlatform":"X","sourceImpressions":38728,"sourceStats":{"asOf":"2026-09-26","likes":413,"reposts":25,"bookmarks":77},"formats":["电影叙事","角色表演"],"hook":{"structure":"静态亮相 → 追打翻腾 → 落船直立收尾","opening":"开场是斗笠橘猫侧坐在青苔石上的近景，嘴里叼根芦苇、背着小武士刀，约 1–2 秒慢慢转头盯向镜头——一只猫摆出浪人架势，第一眼就好笑又带感。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"约 2s 切远景，猫从大石后冲出；约 3–6s 在两个黑衣忍者之间高速穿梭，低机位贴地冲刺、刀光和泥土飞溅。","at":2},{"title":"空中翻腾","text":"约 7–14s 钻进芦苇丛再跃起，仰拍逆天空螺旋翻滚，斗笠甩成圆盘飞离，忍者挥刀扑空。","at":7},{"title":"结尾怎么收","text":"约 18–20s 最后一跳，俯拍落向湖面小船；约 20.7s 一群忍者慢动作扑来，猫落在船头后直立起来，一步步走向镜头推近，身后忍者落水。","at":20.7}],"copyThis":"开头用一个「静止转头盯镜头」的角色亮相，结尾用同样的正面凝视+慢慢推近收，中间才放高速动作。","approx":true},"tags":["30秒 · 武士猫动作","16:9 横屏","Seedance 2.5","TapNow","纯文生视频","芦苇荡忍者"],"steps":[{"number":1,"title":"先写死角色与场景外观","description":"提示词第一段一次性锁定：橘猫白胸白爪、绿金色眼睛，戴编织斗笠，皮带斜背小武士刀，嘴里叼一根干芦苇；场景是起雾的日式湿地、阴天、高高的米色芦苇、青苔石，低饱和大地色、胶片颗粒、浅景深、快速动作带运动模糊。纯文生视频，作者没有用参考图。"},{"number":2,"title":"按段落写动作顺序，每段一个画面任务","description":"依次写：石头上侧脸转头 → 从巨石后冲向镜头 → 两名黑衣忍者间穿梭 → 钻芦苇隧道后高跳旋转 → 空中翻腾与仰拍剪影 → 忍者挥刀、猫空中闪避 → 最后一跳落上小船、忍者慢动作扑来 → 船头低姿态盯镜头后走向镜头。每段一个清楚动作，Seedance 会按顺序剪成快切动作片。"},{"number":3,"title":"最后统一写镜头语言和声音","description":"末段列出机位组合（定机位肖像、低机位跟拍、手持追逐感、低角度、俯拍翻滚、结尾慢推脸）、自然阴天光、雾、湿表面、写实毛发与布料、24fps 胶片感、略去饱和，基调「史诗又有点荒诞」；声音只要风声、芦苇沙沙、水花和远处刀声，无对白。"}],"references_detail":[],"storyboard":[{"number":1,"description":"约 0–2s 开场近景：斗笠橘猫侧坐青苔石上，叼芦苇、背刀，缓缓转头盯镜头。"},{"number":2,"description":"约 2–3s 远景：猫从大青苔石后跃下冲向镜头，芦苇分开。"},{"number":3,"description":"约 3–6.5s 两个黑衣蒙面忍者出现，猫在他们之间高速穿梭，低机位贴地冲过、泥土飞溅。"},{"number":4,"description":"约 6.5–9s 猫压低身子钻进芦苇丛，再冲出跃起，背着刀在灰天前旋转。"},{"number":5,"description":"约 9–13s 空中翻腾、螺旋跳，仰拍剪影；斗笠被甩离，忍者挥刀扑空。"},{"number":6,"description":"约 14–17s 快切：猫在忍者腿间奔跑、滑铲，再次起跳。"},{"number":7,"description":"约 18–22s 最后一跳，俯拍落向湖面小木船；一群忍者从芦苇岸边慢动作跃向小船。"},{"number":8,"description":"约 23–30s 猫落在船头，随后后腿直立沿船中线一步步走向镜头，镜头慢推到脸，身后忍者落水。"}],"constraints":"纯文生视频，无参考图；角色外观一次写全（橘猫白胸白爪绿金眼、编织斗笠、斜背小武士刀、叼芦苇）；阴天雾气湿地、米色芦苇、低饱和、胶片颗粒、运动模糊；每段一个动作按顺序推进；结尾正面凝视+慢推脸；只要风声、芦苇、水花、远处刀声，无对白。缺口：作者未发布参考图（成片截帧不作参考图）；线程内无补充提示词；成片结尾猫为后腿直立行走，提示词未写直立；帖文称是『broken heart cat』系列续集，前作未收录。","video_prompt":{"title":"斗笠武士猫 · Seedance 2.5 提示词","subtitle":"30s · 16:9 · Seedance 2.5 on TapNow · 纯文生视频 · 英文完整提示词（主帖原文）","content":"$4b"}},{"id":"husky-minori-foods-tomato-baton","title":"一颗番茄的接力 · MINORI FOODS 食品企业广告","subtitle":"X · @husky__create · GPT Image 2.5 → Gemini Omni 1.1 Flash · 10秒 · 16:9","description":"GPT Image 2.5 出九宫格分镜、Gemini Omni 成片：一颗番茄从农田接力到餐桌的日本食品企业广告。","video":"https://media.liuyidaoai.com/tutorials/husky-minori-foods-tomato-baton/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/husky-minori-foods-tomato-baton/poster.jpg","duration":"10秒","durationSec":10,"styleLabel":"写实广告","shots":9,"references":1,"model":"GPT Image 2.5 → Gemini Omni 1.1 Flash","style":"日本食品企业广告 · 纪录片式暖调实拍质感","aspectRatio":"16/9","sourceUrl":"https://x.com/husky__create/status/2103422112312541342","sourceAuthor":"@husky__create","sourcePlatform":"X","sourceImpressions":2679,"sourceStats":{"asOf":"2026-09-26","likes":35,"reposts":2,"bookmarks":36},"formats":["产品广告"],"hook":{"structure":"农田采摘 → 运输质检 → 厨房上桌 → 品牌字卡","opening":"第 0 秒是晨光番茄田的大远景，农夫抱着空木箱从两排番茄架中间朝镜头走来，镜头缓慢前推。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"约每 1 秒一镜：约 1s 手摘番茄，约 2s 掌心番茄微距，约 3s 木箱放进白色冷藏车，约 4s 戴白手套在秤上质检，约 5s 厨师切番茄，约 6s 俯拍淋橄榄油的番茄意面。","at":1},{"title":"结尾怎么收","text":"约 7s 一家三口吃饭、女儿咬一口意面；约 8.5s 石桌上的意面和整颗番茄，背景是黄昏农田，左侧淡入「一皿の向こう側。/ MINORI FOODS」。","at":8.5}],"copyThis":"同一颗番茄当「接力棒」串起 9 个 1 秒镜头，每镜只做一个动作，最后留左侧空白放标语。","approx":true},"tags":["10秒 · 企业广告","16:9 横屏","GPT Image 2.5 → Gemini Omni 1.1 Flash","九宫格分镜 → 成片","番茄接力 · 食品农业"],"steps":[{"number":1,"title":"第一步：用 GPT Image 2.5 生成 3×3 九宫格分镜","description":"把「参考图」卡片里的完整分镜提示词（① GPT Image 2.5）贴进 GPT Image 2.5：一张 16:9 图里排 9 格，每格也是 16:9，象牙色细边、左上角白色编号 01–09。提示词写全了农夫、厨师、一家三口的外观，以及 01 晨光农田 → 09 品牌字卡的 9 个画面。"},{"number":2,"title":"第二步：把分镜图 + 视频提示词交给 Gemini Omni 1.1 Flash","description":"上传九宫格分镜，贴下方完整视频提示词（② Gemini Omni 1.1 Flash）。它要求从左到右、从上到下读 Shot 01–09，每格还原成全屏实拍镜头，成片里绝不出现九宫格、边框和编号；镜头之间用干净硬切，禁止人或物变形过渡。"},{"number":3,"title":"第三步：按时间轴锁节奏，声音一起生成","description":"视频提示词给每镜写了精确秒数（0.0–1.1s 农田 … 8.5–10.0s 品牌定帧），每镜只做一个动作；音乐（木吉他 + 钢琴 + 手鼓，约 92 BPM）、环境音、扫码「嘀」声、切菜声，以及约 0.5s 和 6.4s 两句日文女声旁白，都由 Gemini Omni 一并生成。结尾左侧留白放「一皿の向こう側。/ MINORI FOODS」。"}],"references_detail":[{"id":"husky-minori-storyboard-9grid","number":"1","title":"九宫格分镜 · ① GPT Image 2.5 提示词","subtitle":"作者原图 1672×941 · 3×3 镜头 01–09","image":"https://media.liuyidaoai.com/tutorials/husky-minori-foods-tomato-baton/refs/storyboard.jpg","prompt":"$4c"}],"storyboard":[{"number":1,"description":"0–1s 晨光番茄田大远景，农夫抱空木箱走来，缓慢前推。"},{"number":2,"description":"1–2s 中近景：农夫手轻轻拧下一颗红番茄。"},{"number":3,"description":"2–3s 固定微距：番茄躺在农夫掌心，带露珠。"},{"number":4,"description":"3–4s 低角度：农夫把一箱番茄放进白色冷藏车货厢。"},{"number":5,"description":"4–5s 质检车间：白手套转动番茄放上不锈钢秤。"},{"number":6,"description":"5–6s 厨房近景：女厨师在木砧板上切番茄。"},{"number":7,"description":"6–7s 俯拍：橄榄油细流淋在罗勒番茄意面上。"},{"number":8,"description":"7–8.5s 家庭晚餐：女儿吃一口意面，父母相视微笑。"},{"number":9,"description":"8.5–10s 石桌上的意面与整颗番茄，背景黄昏农田，慢推，左侧字卡「一皿の向こう側。/ MINORI FOODS」。"}],"constraints":"九宫格只作参考：成片不得出现网格、边框、编号或拼贴；每格还原为全屏实拍镜头、干净硬切、禁止变形过渡；农夫（约58岁、靛蓝工作衫、米色裤、橄榄绿胶靴）贯穿 01–04，女厨师（约34岁、低马尾、白厨师服）贯穿 06–07；同一颗番茄做视觉锚点；每镜一个动作，禁止急推、快速环绕、延时、漂浮食材；色调叶绿/番茄红/土褐/不锈钢银/暖木/奶白，低饱和；左侧留白放「一皿の向こう側。/ MINORI FOODS」，不加其他文字。缺口：旁白音频未转录核对。","video_prompt":{"title":"② Gemini Omni 1.1 Flash 视频提示词","subtitle":"10s · 16:9 · 上传九宫格分镜作参考 · 英文完整提示词（作者线程第 3 帖）","content":"$4d"}},{"id":"husky-nexarc-construction-brand","title":"从一条线到一座城 · NEXARC 建设公司广告","subtitle":"X · @husky__create · GPT Image 2.5 → Gemini Omni 1.1 Flash · 10秒 · 16:9","description":"GPT Image 2.5 出九宫格分镜、Gemini Omni 成片：从一条蓝色画线到落成街区的日本建筑公司广告。","video":"https://media.liuyidaoai.com/tutorials/husky-nexarc-construction-brand/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/husky-nexarc-construction-brand/poster.jpg","duration":"10秒","durationSec":10,"styleLabel":"企业品牌","shots":9,"references":1,"model":"GPT Image 2.5 → Gemini Omni 1.1 Flash","style":"日本建设企业广告 · 电影感实拍质感","aspectRatio":"16/9","sourceUrl":"https://x.com/husky__create/status/2102337226311180568","sourceAuthor":"@husky__create","sourcePlatform":"X","sourceImpressions":21118,"sourceStats":{"asOf":"2026-09-26","likes":275,"reposts":42,"bookmarks":302},"formats":["产品广告"],"hook":{"structure":"画图 → 施工推进 → 落成街区 → 品牌字卡","opening":"第 0 秒是晨光工作室里的建筑师中景，约 0.4s 切到手握蓝色笔沿尺子画线的特写——从一支笔开场。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"约每 1 秒一镜：约 1.1s 地面上的钴蓝测量线与工人靴子，约 2s 泵车浇筑地基，约 3s 仰拍吊装钢梁，约 4s 幕墙玻璃吸盘安装，约 5s 广场种树铺地。","at":1.1},{"title":"结尾怎么收","text":"约 6s 建筑师推开大堂玻璃门走出，约 7.2s 金色黄昏的街区人来人往，约 8.3s 切到对称蓝调时刻的塔楼夜景，居中白字「NEXARC CONSTRUCTION / まだない景色を、つくる。」。","at":8.3}],"copyThis":"用一条蓝线把「纸上画线」和「工地测量线」硬切对上（match cut），后面按施工顺序每秒一镜推进到落成。","approx":true},"tags":["10秒 · 企业广告","16:9 横屏","GPT Image 2.5 → Gemini Omni 1.1 Flash","九宫格分镜 → 成片","建设施工 · 匹配剪辑"],"steps":[{"number":1,"title":"第一步：用 GPT Image 2.5 生成 3×3 九宫格分镜","description":"把「参考图」卡片里的完整分镜提示词（① GPT Image 2.5）贴进 GPT Image 2.5：一张 16:9 图、9 个等大 16:9 格子、象牙色细边、左上角白色编号 01–09。提示词锁定同一位 36 岁建筑师的外观与安全装备、同一组 18 层主楼 + 两栋 6 层副楼的建筑造型，以及贯穿全片的钴蓝色线条。"},{"number":2,"title":"第二步：把分镜图 + 视频提示词交给 Gemini Omni 1.1 Flash","description":"上传九宫格分镜，贴下方完整视频提示词（② Gemini Omni 1.1 Flash）。它要求逐格还原为全屏实拍镜头，绝不显示九宫格、边框或编号；镜头之间用硬切，只有「纸上画线 → 地面测量线」一处匹配剪辑，禁止人、机械或建筑互相变形。"},{"number":3,"title":"第三步：按时间轴锁节奏，安全与声音一起写","description":"视频提示词给 9 镜写了精确秒数（0.0–1.1s 画线 … 8.5–10.0s 蓝调字卡），并反复强调正确的安全装备与吊装规范；配乐约 104 BPM，8.4s 加日文男声旁白「まだない景色を、つくる。ネクサーク・コンストラクション。」。作者回复说本片是多次抽卡选出的，旁白也不稳定，跟做时多生成几次。"}],"references_detail":[{"id":"husky-nexarc-storyboard-9grid","number":"1","title":"九宫格分镜 · ① GPT Image 2.5 提示词","subtitle":"作者原图 1672×940 · 3×3 镜头 01–09","image":"https://media.liuyidaoai.com/tutorials/husky-nexarc-construction-brand/refs/storyboard.jpg","prompt":"$4e"}],"storyboard":[{"number":1,"description":"0–1.1s 晨光工作室：建筑师中景，切手部特写沿尺子画线。"},{"number":2,"description":"1.1–2.1s 硬切到平整地面上的钴蓝测量线，测量员低机位完成标记。"},{"number":3,"description":"2.1–3.1s 地基大全景：泵车向模板浇筑混凝土。"},{"number":4,"description":"3.1–4.1s 仰拍钢结构：塔吊吊着钢梁缓缓就位。"},{"number":5,"description":"4.1–5.1s 幕墙安装：吸盘吊具把蓝灰玻璃板移入框架，工人系安全绳引导。"},{"number":6,"description":"5.1–6.1s 俯瞰近完工广场：铺地、扶正新种的小树。"},{"number":7,"description":"6.1–7.2s 落成大堂：建筑师推开玻璃门走出，镜头后退跟拍。"},{"number":8,"description":"7.2–8.3s 金色黄昏的街区外景，行人、单车、咖啡座，慢推向主楼。"},{"number":9,"description":"约 8.3–10s 对称蓝调时刻塔楼夜景，居中白字「NEXARC CONSTRUCTION」与「まだない景色を、つくる。」。"}],"constraints":"九宫格只作参考：成片不得出现网格、边框、编号；硬切为主，仅一处画线→测量线匹配剪辑，禁止变形；同一建筑师（36 岁、藏青工作夹克、白衬衫，工地加白色安全帽与反光背心）；同一组 18 层蓝灰玻璃主楼 + 两栋 6 层副楼，造型不得变化；主光始终来自画面左侧，清晨→白天→黄昏→蓝调时刻；正确安全装备与吊装、吊物下无人；字卡居中白字且稳定不逐字动画。缺口：作者自述多次抽卡、旁白不稳定，另可给重机单独参考图（未公开）；音频未转录。","video_prompt":{"title":"② Gemini Omni 1.1 Flash 视频提示词","subtitle":"10s · 16:9 · 24fps · 上传九宫格分镜作参考 · 英文完整提示词（作者线程第 3 帖）","content":"$4f"}},{"id":"husky-nexbuild-city-timelapse-seedance","title":"从空地到一座城 · NEXBUILD 建设延时广告 · Seedance 2.5","subtitle":"X · @husky__create · Seedance 2.5 · BytePlus Lumina · 30秒 · 9:16","description":"Seedance 2.5 竖屏建筑延时广告：同一块空地从夜里放线到高楼林立，最后航拍拉起露出 NEXBUILD。","video":"https://media.liuyidaoai.com/tutorials/husky-nexbuild-city-timelapse-seedance/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/husky-nexbuild-city-timelapse-seedance/poster.jpg","duration":"30秒","durationSec":30,"styleLabel":"写实延时","shots":10,"references":0,"model":"Seedance 2.5（BytePlus Lumina）","style":"建设企业广告 · 固定工地延时 · 写实建筑摄影","aspectRatio":"9/16","sourceUrl":"https://x.com/husky__create/status/2101876852029825290","sourceAuthor":"@husky__create","sourcePlatform":"X","sourceImpressions":8180,"sourceStats":{"asOf":"2026-09-26","likes":52,"reposts":2,"bookmarks":29},"formats":["拆装·制作过程","产品广告"],"hook":{"structure":"空地 → 地基与主体 → 幕墙与街道 → 航拍揭晓品牌","opening":"第 0 秒是天亮前的俯拍空地：围挡、工作灯、成堆钢材，几个小小的工人在裸土上走动——画面几乎是空的，等着被填满。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"机位基本不动：约 3s 挖掘机开挖，约 5–7s 地基和柱子立起、塔吊就位，约 9–12s 钢结构一层层长高，约 13–14s 玻璃幕墙包满整栋楼。","at":3},{"title":"街道和广场成形","text":"约 15–21s 转到楼前：大道从土方到铺沥青画线，再铺草坪、水景和树，傍晚大楼亮灯。","at":15},{"title":"结尾怎么收","text":"约 22s 一名白盔工程师背影仰望大楼，约 25s 镜头拉升成航拍，整片办公街区铺满竖屏，中央浮现「NEXBUILD CORPORATION」。","at":25}],"copyThis":"整条片子锁在同一块地、同一个机位上延时建造，最后才拉起航拍——开头的「空」和结尾的「满」形成对照。","approx":true},"tags":["30秒 · 建设延时","9:16 竖屏","Seedance 2.5","BytePlus Lumina","纯文生视频","空地到城市"],"steps":[{"number":1,"title":"本片没有分镜图：一条长提示词直接出片","description":"作者这条没有用 GPT Image 分镜，只在 BytePlus Lumina 上用 Seedance 2.5 跑一条约 1.2 万字符的英文提示词，直接生成 30 秒 9:16 竖屏成片。提示词先定概念「FROM NOTHING TO A CITY.」，要求整座城从零可见地建起来，建造过程本身就是主角。"},{"number":2,"title":"先锁定同一块工地和空间布局","description":"LOCATION CONTINUITY 一段规定全片只在一块固定工地上、不重置环境：中央是未来广场，左侧中层办公楼群，右侧标志性玻璃塔，背景高层，前景主干道与步行区；最终画面里的每栋楼都必须来自前面出现过的施工区，不允许凭空冒出。"},{"number":3,"title":"按 3 秒一段写 10 个施工阶段，最后才出品牌","description":"时间轴从 0–3s ZERO、3–6s GROUNDWORK 一直写到 27–30s GRAND REVEAL / BRAND，每段写清工种、机械和镜头运动；品牌名 NEXBUILD CORPORATION 要求到最后一场才出现，最后 1.5 秒压暗字标背后的中心区域。"}],"references_detail":[],"storyboard":[{"number":1,"description":"约 0–3s 天亮前俯拍空地：围挡、工作灯、钢材堆，工人放线测量。"},{"number":2,"description":"约 3–5s 挖掘机同时开挖，基坑成形。"},{"number":3,"description":"约 5–8s 地基浇筑、柱子立起，塔吊就位，天色渐亮。"},{"number":4,"description":"约 8–12s 钢结构逐层升高。"},{"number":5,"description":"约 12–15s 玻璃幕墙包覆整栋塔楼。"},{"number":6,"description":"约 15–18s 楼前主干道：土方 → 铺沥青 → 画标线。"},{"number":7,"description":"约 18–21s 广场、草坪、水景和行道树铺设，傍晚大楼亮灯。"},{"number":8,"description":"约 21–24s 白盔工程师背影站在完工广场上仰望大楼。"},{"number":9,"description":"约 24–27s 镜头后退并拉升，街区与行人车流出现。"},{"number":10,"description":"约 27–30s 航拍整片办公街区铺满竖屏，中央字标「NEXBUILD CORPORATION」。"}],"constraints":"30 秒 9:16 竖屏；全片同一块工地、不重置环境，最终所有建筑必须来自先前施工区、不得凭空出现；写实材质与建筑摄影质感，禁卡通、奇幻、玩具感、科幻；品牌名只在最后一场出现，最后 1.5 秒压暗字标背后区域。缺口：本帖无分镜图、无分镜提示词（纯文生视频）；提示词要求的『微缩小人』质感在成片中不明显；成片结尾节奏比提示词时间轴略早；音频未转录。","video_prompt":{"title":"NEXBUILD 从空地到城市 · Seedance 2.5 提示词","subtitle":"30s · 9:16 · Seedance 2.5 on BytePlus Lumina · 纯文生视频 · 英文完整提示词（作者线程第 2 帖）","content":"$50"}},{"id":"garylau-rei-tongue-minimax-h3","title":"绫波丽双次吐舌表情深度驱动 · MiniMax H3","subtitle":"X · @GaryLau0101 · MiniMax H3 ReferenceToVideo · 15秒 · 16:9","description":"MiniMax H3 表情驱动短片：一张参考图锁定绫波丽外观，深度动画带动她两次俏皮吐舌。","video":"https://media.liuyidaoai.com/tutorials/garylau-rei-tongue-minimax-h3/demo-web.mp4","poster":"https://media.liuyidaoai.com/tutorials/garylau-rei-tongue-minimax-h3/poster.jpg","duration":"15秒","durationSec":15,"styleLabel":"真人风","shots":1,"references":5,"model":"MiniMax H3（ReferenceToVideo）","style":"深度动画驱动表情 · 固定机位写实中近景","aspectRatio":"16/9","sourceUrl":"https://x.com/GaryLau0101/status/2103326557783925108","sourceAuthor":"@GaryLau0101","sourcePlatform":"X","sourceImpressions":2272,"sourceStats":{"asOf":"2026-09-25","likes":53,"reposts":4,"bookmarks":66},"formats":["角色表演"],"hook":{"structure":"固定机位 · 两次吐舌表情","opening":"白墙前的自拍式正面近景，黑长直女孩穿浅蓝针织衫直视镜头，约 3s 就眨眼吐舌——第一个表情来得很快。","openingAt":0,"beats":[{"title":"过程怎么推进","text":"吐舌后闭眼说话，约 7s 抬手指戳脸颊，约 8–9s 手指点脸再吐一次舌，比第一次更夸张。","at":3},{"title":"结尾怎么收","text":"约 11s 手比「一点点」，之后放松微笑对镜头说话，没有额外剪辑。","at":11}],"copyThis":"机位和构图全程不动，只靠脸部表情撑住：两次吐舌卡在约 3s 和约 8.5s。","approx":true},"tags":["15秒 · 表情驱动","16:9 横屏","MiniMax H3","ReferenceToVideo","深度动画驱动","双次吐舌"],"steps":[{"number":1,"title":"先锁定身份与环境，只让 Picture 1 决定外观","description":"提示词的 subject_definitions 把（绫波丽：同一张脸、自然肤质、深色眼睛、黑长直齐刘海、浅蓝圆领针织衫、银项链）和（白色纹理墙与沙发边）都绑定到

每一个大师,都从小白开始

蒙娜丽莎搬进街坊 · Flova 分屏教程

奢侈沙发拆装工坊 Reels · Seedance 2.0

威尼斯折叠成球 · 粉发女孩送信

蒙娜丽莎搬进街坊 · Flova 分屏教程

奢侈沙发拆装工坊 Reels · Seedance 2.0

威尼斯折叠成球 · 粉发女孩送信

按套路挑

桌面小滑板女孩 · 2D 手绘混实拍 · Seedance 2.530 秒2D 混实拍短片，怎么做出来的？

灰姑娘阁楼晨间 vlog · 4 张设定图锁一致 · Seedance 2.530 秒童话真人 vlog 短片，怎么做出来的？

Therabody 眼部按摩仪广告 · 主持 + 演员双角色 · Seedance 2.540 秒产品广告短片，怎么做出来的？

虚拟穿搭 OOTD · 格纹上衣 + 西部靴 · Seedance 2.520 秒时尚展示短片，怎么做出来的？

1890 年代 GRWM · 束腰到骑车出门 · Seedance 2.530 秒古装真人短片，怎么做出来的？

瑜伽护腕支撑块广告 · 两集 14 镜 · Seedance 2.540 秒写实广告短片，怎么做出来的？

5 张参考图同时锁定 · 维生素软糖两集广告 · Seedance 2.540 秒写实广告短片，怎么做出来的？

乡村农场晨间 vlog · 男主 + 狗 · Seedance 2.530 秒写实生活短片，怎么做出来的？

90 年代超模美妆广告 · GRWM 30 镜变装 · Seedance 2.530 秒时尚广告短片，怎么做出来的？

走廊跳舞 + 小朋友手机跟拍 · 18 格故事板复刻 · Seedance 2.512 秒3D 卡通短片，怎么做出来的？

地铁通道以一敌多 · 故事板驱动动作戏 · Seedance 2.530 秒写实动作短片，怎么做出来的？

香料米饭烹饪 vlog · 6 段场景 + 英文口播 · Seedance 2.520 秒美食 vlog 短片，怎么做出来的？

奶油鸡肉意面烹饪 vlog · 角色设定图 + 6 图参考 · Seedance 2.530 秒美食 vlog 短片，怎么做出来的？

雨夜入侵者对峙 · 好莱坞惊悚短片 · Seedance 2.530 秒电影感短片，怎么做出来的？

ONE HIT 时尚造型图册 MV · 1.5 秒一换装卡点 · Seedance 2.515 秒时尚变装短片，怎么做出来的？

骑车一脚踹飞重卡 · 第一视角反物理6 秒第一视角整活短片，怎么做出来的？

挠痒高跟鞋 · 白模视频驱动试穿 · Seedance 2.515 秒白模转写实短片，怎么做出来的？

口袋云台相机开箱 · 创作者向评测30 秒科技开箱短片，怎么做出来的？

手一扫就变脸 · 双人自拍整活10 秒自拍整活短片，怎么做出来的？

暗调汉堡广告 · 14 镜头快切15 秒美食广告短片，怎么做出来的？

哈利法塔重力切换建造 · FPV 一镜到底15 秒建造过程短片，怎么做出来的？

暴躁怪兽做南瓜派 · 万圣节美食 vlog14 秒美食 vlog 短片，怎么做出来的？

认错王子 · 卡丁车抢人反转15 秒喜剧反转短片，怎么做出来的？

管道女孩 · 浮空王国跑酷15 秒游戏CG风短片，怎么做出来的？

蒙面女侠 · 酒馆打斗跳马追马32 秒复古喜剧短片，怎么做出来的？

蒙面女侠 · 酒馆门廊出糗19 秒复古喜剧短片，怎么做出来的？

麂皮手提包 · UGC 开箱口播20 秒真人风短片，怎么做出来的？

粉色夜城说唱 MV · MiniMax H312 秒真人风短片，怎么做出来的？

棕色乐福鞋 · 开箱试穿种草 · Seedance 2.523 秒真人风短片，怎么做出来的？

蒸汽朋克飞艇酒吧 · 小猴帮打 · Seedance 2.520 秒真人风短片，怎么做出来的？

黑色束腰机库大片 · VHS 复古手持 · Seedance 2.024 秒真人风短片，怎么做出来的？

植物大战僵尸 · 向日葵的救赎 · Seedance 2.530 秒真人风短片，怎么做出来的？

报复性熬夜 · 黑眼圈越刷越深 · Seedance 2.530 秒写实生活短片，怎么做出来的？

别惹 THAROG · 猎人射兽逃进山洞 · Seedance 2.530 秒史诗写实短片，怎么做出来的？

我怎么上广告牌了 · 涩谷夜景海报眨眼15 秒时尚广告短片，怎么做出来的？

复制粘贴就能做的动态图形片头 · MiniMax H315 秒扁平动态图形短片，怎么做出来的？

微缩城市的戴冠式 · 巨幅公主肖像揭幕15 秒微缩模型短片，怎么做出来的？

一滴辣酱，全员警报 · MAYDAY 辣酱广告30 秒写实广告短片，怎么做出来的？

博物馆台阶上的静止人群 · Seedance 2.530 秒超现实短片，怎么做出来的？

红发双刀 vs 单刀武士 · 云端空战剑斗 · Seedance 2.530 秒手绘动漫短片，怎么做出来的？

斗笠武士橘猫 · 芦苇荡斗忍者 · Seedance 2.530 秒写实动作短片，怎么做出来的？

一颗番茄的接力 · MINORI FOODS 食品企业广告10 秒写实广告短片，怎么做出来的？

从一条线到一座城 · NEXARC 建设公司广告10 秒企业品牌短片，怎么做出来的？

从空地到一座城 · NEXBUILD 建设延时广告 · Seedance 2.530 秒写实延时短片，怎么做出来的？

绫波丽双次吐舌表情深度驱动 · MiniMax H315 秒真人风短片，怎么做出来的？

漫画本脱出剧 · MiniMax H315 秒真人×漫画短片，怎么做出来的？

前排女孩与歌手的瞬间对视 · Seedance 2.530 秒手机实拍短片，怎么做出来的？

白模绿幕抠像 · 动作复刻 · MiniMax H38 秒白模动画短片，怎么做出来的？

鎏金巴洛克聚光高定十姿 · Seedance 2.515 秒时装摄影短片，怎么做出来的？

未完工空间的折纸焕新 · Gemini Omni10 秒写实室内短片，怎么做出来的？

绛红手机的一日漫游广告 · Seedance 2.530 秒写实广告短片，怎么做出来的？

古装红纱闺房 · 六镜眼神戏 · Seedance 2.530 秒古装真人短片，怎么做出来的？

字筑暴走 · 二次元时装动作 · MiniMax H315 秒二次元短片，怎么做出来的？

二次元动态片头 · 字效踩点 · MiniMax H315 秒二次元短片，怎么做出来的？

飞毯穿城 · 史诗幻想航拍 · 15s15 秒奇幻航拍短片，怎么做出来的？

走廊蓝蝶化身 · 红眼亮灯出场 · Seedance 2.516 秒暗黑写实短片，怎么做出来的？

特洛伊战地 Vlog · 2026 穿越 · Seedance 2.530 秒手机 Vlog 短片，怎么做出来的？

韩风穿搭 Lookbook 转盘 · Seedance 2.517 秒时尚展示短片，怎么做出来的？

车内密戏眼神拉扯 · Seedance 2.522 秒真人风短片，怎么做出来的？

宝马变越野 · 汽修 Reels · Seedance 2.010 秒写实广告短片，怎么做出来的？

陶阿狗色卡变装 · 行者AI拆解11 秒时尚变装短片，怎么做出来的？

我们之间的路——暴风雨中相遇的两个陌生人30 秒电影感短片，怎么做出来的？

把海洋和天空折叠成一条毯子15 秒超现实短片，怎么做出来的？

奢侈沙发拆装工坊 Reels · Seedance 2.010 秒产品 CGI 短片，怎么做出来的？

潮汐水字 TIDAL · Minimax H3 流体字效15 秒流体特效短片，怎么做出来的？

时尚编辑板分镜 · MiniMax H3（1:1）15 秒时尚编辑短片，怎么做出来的？

MacBook Pro 开箱 · 创作者向评测15 秒科技开箱短片，怎么做出来的？

1907 印度村落打水 · 仿古默片15 秒黑白默片短片，怎么做出来的？

1906 巴黎早市 · 仿古默片15 秒黑白默片短片，怎么做出来的？

Sony 旗舰耳机开箱 · YouTube Shorts25 秒科技开箱短片，怎么做出来的？

十姿态时尚闪拍 · 角色一致性15 秒时装摄影短片，怎么做出来的？

敦煌群舞白膜成片 · MiniMax Design 全流程20 秒古风写实短片，怎么做出来的？

每一个大师,
都从小白开始

桌面小滑板女孩 · 2D 手绘混实拍 · Seedance 2.5
30 秒2D 混实拍短片，怎么做出来的？

灰姑娘阁楼晨间 vlog · 4 张设定图锁一致 · Seedance 2.5
30 秒童话真人 vlog 短片，怎么做出来的？

Therabody 眼部按摩仪广告 · 主持 + 演员双角色 · Seedance 2.5
40 秒产品广告短片，怎么做出来的？

虚拟穿搭 OOTD · 格纹上衣 + 西部靴 · Seedance 2.5
20 秒时尚展示短片，怎么做出来的？

1890 年代 GRWM · 束腰到骑车出门 · Seedance 2.5
30 秒古装真人短片，怎么做出来的？

瑜伽护腕支撑块广告 · 两集 14 镜 · Seedance 2.5
40 秒写实广告短片，怎么做出来的？

5 张参考图同时锁定 · 维生素软糖两集广告 · Seedance 2.5
40 秒写实广告短片，怎么做出来的？

乡村农场晨间 vlog · 男主 + 狗 · Seedance 2.5
30 秒写实生活短片，怎么做出来的？

90 年代超模美妆广告 · GRWM 30 镜变装 · Seedance 2.5
30 秒时尚广告短片，怎么做出来的？

走廊跳舞 + 小朋友手机跟拍 · 18 格故事板复刻 · Seedance 2.5
12 秒3D 卡通短片，怎么做出来的？

地铁通道以一敌多 · 故事板驱动动作戏 · Seedance 2.5
30 秒写实动作短片，怎么做出来的？

香料米饭烹饪 vlog · 6 段场景 + 英文口播 · Seedance 2.5
20 秒美食 vlog 短片，怎么做出来的？

奶油鸡肉意面烹饪 vlog · 角色设定图 + 6 图参考 · Seedance 2.5
30 秒美食 vlog 短片，怎么做出来的？

雨夜入侵者对峙 · 好莱坞惊悚短片 · Seedance 2.5
30 秒电影感短片，怎么做出来的？

ONE HIT 时尚造型图册 MV · 1.5 秒一换装卡点 · Seedance 2.5
15 秒时尚变装短片，怎么做出来的？

骑车一脚踹飞重卡 · 第一视角反物理
6 秒第一视角整活短片，怎么做出来的？

挠痒高跟鞋 · 白模视频驱动试穿 · Seedance 2.5
15 秒白模转写实短片，怎么做出来的？

口袋云台相机开箱 · 创作者向评测
30 秒科技开箱短片，怎么做出来的？

手一扫就变脸 · 双人自拍整活
10 秒自拍整活短片，怎么做出来的？

暗调汉堡广告 · 14 镜头快切
15 秒美食广告短片，怎么做出来的？

哈利法塔重力切换建造 · FPV 一镜到底
15 秒建造过程短片，怎么做出来的？

暴躁怪兽做南瓜派 · 万圣节美食 vlog
14 秒美食 vlog 短片，怎么做出来的？

认错王子 · 卡丁车抢人反转
15 秒喜剧反转短片，怎么做出来的？

管道女孩 · 浮空王国跑酷
15 秒游戏CG风短片，怎么做出来的？

蒙面女侠 · 酒馆打斗跳马追马
32 秒复古喜剧短片，怎么做出来的？

蒙面女侠 · 酒馆门廊出糗
19 秒复古喜剧短片，怎么做出来的？

麂皮手提包 · UGC 开箱口播
20 秒真人风短片，怎么做出来的？

粉色夜城说唱 MV · MiniMax H3
12 秒真人风短片，怎么做出来的？

棕色乐福鞋 · 开箱试穿种草 · Seedance 2.5
23 秒真人风短片，怎么做出来的？

蒸汽朋克飞艇酒吧 · 小猴帮打 · Seedance 2.5
20 秒真人风短片，怎么做出来的？

黑色束腰机库大片 · VHS 复古手持 · Seedance 2.0
24 秒真人风短片，怎么做出来的？

植物大战僵尸 · 向日葵的救赎 · Seedance 2.5
30 秒真人风短片，怎么做出来的？

报复性熬夜 · 黑眼圈越刷越深 · Seedance 2.5
30 秒写实生活短片，怎么做出来的？

别惹 THAROG · 猎人射兽逃进山洞 · Seedance 2.5
30 秒史诗写实短片，怎么做出来的？

我怎么上广告牌了 · 涩谷夜景海报眨眼
15 秒时尚广告短片，怎么做出来的？

复制粘贴就能做的动态图形片头 · MiniMax H3
15 秒扁平动态图形短片，怎么做出来的？

微缩城市的戴冠式 · 巨幅公主肖像揭幕
15 秒微缩模型短片，怎么做出来的？

一滴辣酱，全员警报 · MAYDAY 辣酱广告
30 秒写实广告短片，怎么做出来的？

博物馆台阶上的静止人群 · Seedance 2.5
30 秒超现实短片，怎么做出来的？

红发双刀 vs 单刀武士 · 云端空战剑斗 · Seedance 2.5
30 秒手绘动漫短片，怎么做出来的？

斗笠武士橘猫 · 芦苇荡斗忍者 · Seedance 2.5
30 秒写实动作短片，怎么做出来的？

一颗番茄的接力 · MINORI FOODS 食品企业广告
10 秒写实广告短片，怎么做出来的？

从一条线到一座城 · NEXARC 建设公司广告
10 秒企业品牌短片，怎么做出来的？

从空地到一座城 · NEXBUILD 建设延时广告 · Seedance 2.5
30 秒写实延时短片，怎么做出来的？

绫波丽双次吐舌表情深度驱动 · MiniMax H3
15 秒真人风短片，怎么做出来的？

漫画本脱出剧 · MiniMax H3
15 秒真人×漫画短片，怎么做出来的？

前排女孩与歌手的瞬间对视 · Seedance 2.5
30 秒手机实拍短片，怎么做出来的？

白模绿幕抠像 · 动作复刻 · MiniMax H3
8 秒白模动画短片，怎么做出来的？

鎏金巴洛克聚光高定十姿 · Seedance 2.5
15 秒时装摄影短片，怎么做出来的？

未完工空间的折纸焕新 · Gemini Omni
10 秒写实室内短片，怎么做出来的？

绛红手机的一日漫游广告 · Seedance 2.5
30 秒写实广告短片，怎么做出来的？

古装红纱闺房 · 六镜眼神戏 · Seedance 2.5
30 秒古装真人短片，怎么做出来的？

字筑暴走 · 二次元时装动作 · MiniMax H3
15 秒二次元短片，怎么做出来的？

二次元动态片头 · 字效踩点 · MiniMax H3
15 秒二次元短片，怎么做出来的？

飞毯穿城 · 史诗幻想航拍 · 15s
15 秒奇幻航拍短片，怎么做出来的？

走廊蓝蝶化身 · 红眼亮灯出场 · Seedance 2.5
16 秒暗黑写实短片，怎么做出来的？

特洛伊战地 Vlog · 2026 穿越 · Seedance 2.5
30 秒手机 Vlog 短片，怎么做出来的？

韩风穿搭 Lookbook 转盘 · Seedance 2.5
17 秒时尚展示短片，怎么做出来的？

车内密戏眼神拉扯 · Seedance 2.5
22 秒真人风短片，怎么做出来的？

宝马变越野 · 汽修 Reels · Seedance 2.0
10 秒写实广告短片，怎么做出来的？

陶阿狗色卡变装 · 行者AI拆解
11 秒时尚变装短片，怎么做出来的？

我们之间的路——暴风雨中相遇的两个陌生人
30 秒电影感短片，怎么做出来的？

把海洋和天空折叠成一条毯子
15 秒超现实短片，怎么做出来的？

奢侈沙发拆装工坊 Reels · Seedance 2.0
10 秒产品 CGI 短片，怎么做出来的？

潮汐水字 TIDAL · Minimax H3 流体字效
15 秒流体特效短片，怎么做出来的？

时尚编辑板分镜 · MiniMax H3（1:1）
15 秒时尚编辑短片，怎么做出来的？

MacBook Pro 开箱 · 创作者向评测
15 秒科技开箱短片，怎么做出来的？

1907 印度村落打水 · 仿古默片
15 秒黑白默片短片，怎么做出来的？

1906 巴黎早市 · 仿古默片
15 秒黑白默片短片，怎么做出来的？

Sony 旗舰耳机开箱 · YouTube Shorts
25 秒科技开箱短片，怎么做出来的？

十姿态时尚闪拍 · 角色一致性
15 秒时装摄影短片，怎么做出来的？

敦煌群舞白膜成片 · MiniMax Design 全流程
20 秒古风写实短片，怎么做出来的？

蒙娜丽莎搬进街坊 · Flova 分屏教程
60 秒超现实短片，怎么做出来的？

威尼斯折叠成球 · 粉发女孩送信
30 秒复古胶片短片，怎么做出来的？

刺绣缝 Logo · 线自己爬上织物
11 秒刺绣微距短片，怎么做出来的？

红绳拼字 Smiling · 定格小人工坊
10 秒定格动画短片，怎么做出来的？

法属波利尼西亚冒险 · 碎切度假记忆
30 秒手持实拍短片，怎么做出来的？

ROOM 214 · H3 定格收拾行李
15 秒定格动画短片，怎么做出来的？

敦煌飞天桌面壁纸翻车 · H3
10 秒桌面动画短片，怎么做出来的？