Промпты Seedance 2.5: 1 630+ примеров и руководство

EMBER FRACTURE — AWAKENING Image SO-YEON | 30s | 16:9 | one continuous generation. LIVE-ACTION ONLY: Ultra-photorealistic AAA supernatural martial arts. <<<image_1>>> locks SO-YEON's exact identity; same single woman throughout. Small 160cm build, messy black hair, black hoodie/pants/Converse. Real skin, pores, hair, eyes and biomechanics. Never CGI/digital double. ALLEY: Wet narrow convenience-store alley at night, red/pink/teal neon, puddles, crates, dumpster, vending machine, steam. 5 unarmed male delinquents attack aggressively, never taking turns. AUDIO: Korean dialogue, diegetic impacts/traffic/ambience only. No BGM/subtitles. AWAKENED COMBAT: Her body takes over automatically; unarmed Taekwondo × Karate × Pencak Silat, extremely fast, fluid chained movement with no reset poses. VFX LOCK: Acceleration = physical movement + subtle heat-shimmer/refractive distortion only. No fire, flames, embers, orange particles, beams, plasma or teleportation. Every strong physical impact creates a brief glass-like spiderweb fracture in the air, splitting neon light, then silently disappearing; realistic steam/dust/debris from impact. IMPACT → FRACTURE WEB → NEON REFRACTION → DISSIPATION. 00–03: So-Yeon cowers against wall. Leader smacks her head. She slowly glares through bangs. {어쭈? 이 년 봐라? 많이 컸다?} 03–06: Fear returns; she flinches: {아…! 미안…}. Leader swings again. 06–09: 0.3s slow-mo: automatic head-slip → Silat parry → pivot → body punch. Fracture flash. Leader crashes into crates. 09–12: She stares at trembling hand in disbelief. {저 년이?!} Four rush simultaneously. 12–17: Heat-shimmer dash through two: parry → redirect → elbow → sweep → side kick. Impact fracture + steam. Continue instantly. 17–22: Two attack together: diagonal dash → duck → shoulder redirect → roundhouse. Wide fracture-web splits pink/teal neon. 22–27: Final attacker from behind. Maximum physical dash; camera lags/whip-pans. Evade → parry → jab → body punch → low kick → spinning back kick. 0.25s slow-mo; dense fracture-web across wall. 27–30: Largest fracture lattice spreads across alley, neon reflections split, debris/crates fly, enemy slides away. Fracture hits lens → brief refracted distortion. So-Yeon studies her hands, then calmly walks through steam, half frightened/half amazed, hair falling over eyes. Never looks back. CUT. CAMERA: Close, kinetic handheld pursuit; tracking, whip-pans, low angles, OTS, facial impacts, rapid focus. Camera reacts slightly late to acceleration. Impact shake only on physical contact. Readable action > chaos. HARD LOCK: Real human + real environment + VFX in same frame. No duplicate So-Yeon. No teleportation. Every VFX effect must originate from visible physical contact.

Went out for milk, came back barefoot 😭 GENERATED With Seedance 2.5 Prompt: 👇 Create a 30-second ultra-realistic Korean summer-morning vlog set in South Korea in the early 2000s, filmed casually by a friend on an authentic consumer DV camcorder. It must feel like genuine home-video footage, not a modern video with a retro filter. CHARACTER: The same young Korean woman in her early 20s throughout. Natural appearance, black wavy hair in a loose messy side ponytail, fitted pastel-blue short top, loose white pajama pants, simple silver necklace, casual flat sandals. Realistic skin texture, minimal makeup, natural expressions and movement. LOCATION: A peaceful Korean residential neighborhood on a warm early-summer morning. Narrow concrete lane, small old houses, weathered concrete walls, potted plants, bicycles, utility poles, overhead wires, laundry hanging outside, and a small neighborhood convenience store. CRITICAL SANDAL RULE (highest priority): From 0s until 17s, BOTH sandals are worn on her feet, properly on both feet, and she holds NO footwear in either hand at any point. Her only hand-held item before 17s is the cloth shopping bag (and later the milk inside it). The sandals come off her feet ONLY at the end of the FINAL ATTEMPT scene (around 21–23s). Before that moment, no sandal may appear in her hand, under her arm, or anywhere except on her feet. EXACT ORDER, 6 SCENES: 0–4s, LEAVING HOME: She steps out of her house wearing both sandals on her feet, holding only a small cloth shopping bag. She locks the door and starts walking down the narrow lane. Both feet in sandals, hands free of footwear. 4–8s, SCHOOL CHILDREN: A few school children with backpacks walk past her. She smiles, waves and says "Good morning!" The children wave back happily and continue ahead toward school. She is still wearing both sandals normally. 8–12s, STORE: She enters the small convenience store (door chime), buys one carton of milk, puts it carefully in her bag and walks back out. Both sandals still on her feet, nothing in her hands except the bag. 12–17s, SANDAL TROUBLE: Walking home through the same lane, she suddenly stops and looks down. The strap of her right sandal has come loose, but the sandal is still on her foot. She pushes the strap back with her fingers and it slips out again. She crouches and tries another way to fix it while the sandal stays on her foot, but it won't stay. 17–23s, FINAL ATTEMPT: The friend zooms in awkwardly. She laughs, makes one last attempt, and the strap pops loose again. Only now, she gives up, takes off both sandals and holds them together in one hand, with the milk bag in the other hand. This is the FIRST moment any sandal is in her hand. 23–30s, BAREFOOT WALK: She walks barefoot down the sunny lane, sandals in one hand, milk bag in the other. The same school children, now farther ahead, notice her sandals and smile, and she laughs and waves. She turns to the camera and says, "I only went out for milk." She keeps w

30 sec | 16:9 | One continuous generation | English dialogue | No BGM | No subtitles STYLE: Ultra-photorealistic live-action premium smartphone commercial with the cinematic visual language of a coming-of-age movie. Real actress, real environment, natural practical lighting, subtle 35mm film grain, realistic skin texture, physically accurate reflections, natural handheld imperfections. The first half must feel like a movie scene, while the product is gradually revealed as the reason the moment exists. CHARACTER LOCK: YOUNG WOMAN from the provided reference image. Same actress throughout. Long messy wavy dark-brown hair, natural skin texture, grey fitted cropped T-shirt, black high-waisted jeans. Preserve her exact facial appearance, hairstyle, clothing, body proportions and styling from the reference image. Natural micro-expressions, subtle acting, realistic movement. PRODUCT LOCK: VELORA V1, fictional premium smartphone based exactly on the provided phone reference image. Deep burgundy metallic body, ultra-thin flat frame, minimal black dual-camera module, subtle silver "V" logo centered on the back, premium glass finish. Keep the exact same phone design, proportions, color, camera placement and logo throughout the entire video. No real-world smartphone branding. LOCATION: Outdoor rooftop terrace at blue hour. Deep navy evening sky with a faint warm pink horizon, distant trees and soft city lights. Weathered concrete ledge. Gentle natural wind. Quiet cinematic atmosphere. SOUND: Natural location sound only. Evening wind, distant traffic, footsteps, fabric movement, subtle breathing, realistic phone camera sounds and interface clicks. No music. STORY: The commercial begins as if it is a scene from a real movie. The audience does not immediately realize they are watching a smartphone advertisement. 00:00–05:00 — SETTING THE CAMERA Extreme close-up of a hand carrying the burgundy VELORA V1. The hand places the phone carefully on a concrete ledge. Close-up of the phone's burgundy metallic back catching the blue-hour light. The girl reaches toward the phone. She taps the screen. The front camera opens. The phone screen shows her live preview. She adjusts the phone angle slightly until the composition matches the provided girl reference image. She steps backward. A small realistic recording indicator appears: "REC 00:00" Camera remains fixed as if the phone itself is recording. 05:00–11:00 — THE REFERENCE PHOTO COMES ALIVE The recorded footage now becomes the main visual. Recreate the exact composition and appearance of the provided girl reference image: Low-angle perspective. Deep blue evening sky. Long messy wavy hair. Grey cropped T-shirt. Black jeans. Natural direct-camera expression. She stands still for a moment. A gentle gust of wind moves her hair naturally across her face. She looks directly into the front camera. She gives a tiny genuine smile. She quietly says: "Ok

CAMERA/LOOK: DV 16mm tape camcorder POV, handheld/propped shots. Shaky framing, imperfect focus, clumsy zooms, occasional face cut-offs. Soft blurry tape quality, faint noise, bloomed gym lights, flickering exposure, muted contrast, realistic skin tones. Camcorder never visible. STYLE: Playful, self-deprecating gym vlog. Genuine strain, humor, quick energetic cuts, lighthearted rather than exhausted. CHARACTER: CHASE, Korean idol in her 20s. Long black high ponytail, glowing slightly sweaty skin, expressive eyes, slim athletic build. Modest long-sleeve athletic top, joggers/leggings, socks, no jewelry. SETTING: Evening gym mat area, nearby mirrors, water bottle on floor, soft overhead lighting, equipment in background. 15s / 6 CUTS: 1. (~2.5s) Propped medium shot. CHASE lies on mat, propped on elbows, sighing dramatically. "Okay, core day — my least favorite day." 2. (~2.5s) Plank, slight arm shaking, strained expression. "Why does this get harder every single time—" 3. (~2.5s) Crunches, breathing hard and complaining between reps. "Nobody warns you how much this actually burns." 4. (~2s) Macro close-up, shallow DOF. Hands grip mat during leg raises, abs visibly working. No dialogue, gym ambience. 5. (~2.5s) Handheld/propped shot. Finishes leg raises, flops onto mat laughing. "Okay— that's it, I'm done—" 6. (~3s) Arm's-length selfie while lying on mat, tired grin. "Core day never gets easier — see you guys tomorrow, hopefully."

Photorealistic live-action video, real-time speed throughout. NO slow motion. Refer to Image 1. Inside a 3D anamorphic billboard on a modern building, an elegant woman with wavy hair wears a red high-slit velvet gown and reclines sideways on a marble ledge. A red crocodile-skin handbag is beside her. She quickly grabs the handbag, lifts it, then forcefully SLINGS it forward toward the viewer. Her arm makes one fast forward swing and releases the bag with strong force. The moment the bag leaves her hand, it must immediately shoot rapidly toward the foreground hand. The flight is FAST and SHORT, with no lingering airborne time. Do not show the bag slowly traveling through the air. It moves quickly from the woman's hand directly toward the catcher. The bag has real weight and momentum. Gravity naturally pulls it slightly downward during the fast flight, but it must continue moving rapidly forward. No floating, hovering, slow drifting, or suspended motion. A manicured foreground hand catches the SAME handbag almost immediately after the throw. The hand and wrist recoil naturally from the bag's momentum and weight. ONE continuous shot. No cuts, hidden transitions, camera resets, or duplicated objects. Exactly ONE handbag from beginning to end. The bag leaves the woman's hand once and is caught once. Never duplicate, replace, regenerate, or recreate the bag. Preserve the woman's face, hair, gown, handbag, marble ledge, billboard and building. Realistic physics, anatomy and motion. No morphing, flickering or teleportation. NO SLOW MOTION AT ANY POINT. No artificial SFX. Natural ambient sound with relaxed, sensual background music.

16:9 横屏，30秒，电影级现代夜间特种作战短片，超写实真人摄影，高预算军事惊悚片质感。整体色调为冷蓝黑色，低饱和、低照度、高反差，空气中有轻微雾气、灰尘和冷凝水汽。场景为深夜城市高层建筑内部及相连的工业走廊，包含狭窄通道、金属门、混凝土墙面、低亮度应急灯和窗外模糊城市灯光。全片保持同一夜间环境、同一人物、同一服装、同一枪械、同一冷蓝光线，不改变时间、地点和视觉风格。 主角高度参考上传图片中的同一名特种兵。深色夜战作战服，宽松深色战术外套，深色贴头帽或头套，黑色战术手套，胸前有软管式呼吸或通信装置，肩背黑色战术装备包。主角手持同一把现代突击步枪，枪械安装消音器、瞄准镜、战术灯和前握把。人物始终保持冷静、克制、专业、专注。面部、服装、装备和枪械在整段视频中保持一致，不换脸、不换装、不换武器。 0至5秒： 以接近上传图片气质的中近景开始。摄影机位于主角正前方略低位置，35mm镜头，轻微仰拍。主角单膝或低姿停在黑暗走廊边缘，低头快速确认武器状态，然后缓慢抬头观察前方。冷蓝色侧光打在脸部、帽子和武器边缘，背景大部分沉入黑暗。窗外只有模糊蓝白城市灯点。摄影机缓慢向前推进，不做快速切换。声音只保留沉重呼吸、衣料摩擦、远处风声和轻微无线电杂音。 5至10秒： 主角起身，以低姿贴着右侧墙面向前推进。枪口始终朝向前方，动作安静、缓慢、明确。摄影机保持在主角左后方肩后位置，以35mm镜头稳定跟随。主角经过一个转角时短暂停顿，身体贴墙，先观察，再继续移动。前方远处出现一名守卫的模糊身影。守卫缓慢转身巡视，不要突然出现。空间关系保持清晰，主角在近景，守卫在走廊远端。 10至15秒： 主角接近守卫所在区域，停在门框阴影内。守卫开始察觉异常，头部轻微转向主角方向。主角立即抬枪，用消音武器完成一次极短促点射。守卫产生自然受击反应并倒入墙边或地面，不出现夸张血腥。主角不原地停留，立刻跨过守卫位置继续前进。摄影机仍保持肩后跟拍，不切成太多角度。重点表现“发现、反应、压制、继续推进”的清晰动作链。 15至20秒： 主角来到目标房间门外。房门关闭或半掩，门缝中透出微弱蓝白光。主角靠墙站位，短暂停顿，调整握枪姿势，侧头确认室内动静。摄影机从肩后缓慢移动到主角侧前方，保持同一空间连续性。主角迅速推门进入，不使用复杂爆破，不增加多人协同。房间内只有一名高价值目标和一名贴身护卫。护卫位于目标侧后方，目标靠近桌面、设备或窗边。 20至25秒： 主角进入房间后立即控制射界。护卫首先发现主角，试图抬起武器。主角迅速进行一次短点射压制护卫，然后立即将枪口转向高价值目标。高价值目标后退一步，试图躲向桌边或墙角。主角保持稳定站姿和精准持枪动作，迅速完成目标清除。整个过程只安排两次明确动作：先压制护卫，再锁定目标。避免同时多人乱动，避免复杂枪战。不要血浆特写，不要慢动作，不要大规模爆炸。 25至30秒： 主角快速扫视房间确认安全，低头看一眼胸前通讯装置，短暂停顿，随后转身撤离。摄影机跟随主角从房门退出，再次回到冷蓝色走廊。最后3秒使用与开头相似的低机位中近景。主角持枪向前进入更深的黑暗，冷蓝色雾气和远处城市灯光出现在背景。主角轮廓逐渐被黑暗吞没，画面在沉重呼吸声和微弱无线电杂音中切黑。 Seedance 2.5 稳定执行规则： 人物一致性为最高优先级。主角全程保持同一张脸、同一深色头套、同一作战服、同一战术外套、同一胸前软管装置、同一背包和同一把步枪。不要生成额外队员，不要生成大量敌人。全片最多出现主角、一名走廊守卫、一名贴身护卫和一名高价值目标。 镜头数量控制在少量长镜头和清晰过渡。避免连续快速剪辑，避免360度绕拍，避免突然切到完全不同空间。摄影机以35mm为主，必要时使用50mm近景。主要采用肩后跟拍、侧前方中近景和轻微低机位。摄影机运动平稳，保留轻微手持呼吸感。 动作必须简洁明确。每个时间段只执行一个主要动作目标。不要同时安排大量奔跑、多人交火、爆炸、直升机、烟雾和快速切镜。优先保证人物、武器、动作和空间连续性。 所有枪械保持相同结构、长度和附件位置。枪托始终贴合肩部，双手持枪姿态符合真实人体结构。禁止枪管弯曲、弹匣变形、瞄具漂移、武器忽然变长或变短。 夜间光线全程保持冷蓝黑色调。禁止突然出现暖色日光、明亮白天、红色霓虹或赛博朋克环境。暗部保留细节，皮肤、衣料、金属和墙体要有真实纹理。 负面约束： 人物换脸，人物变装，突然换枪，武器变形，枪管弯曲，瞄具漂移，弹匣消失，手指穿模，多余手指，多余肢体，人物瞬移，场景跳变，大量群演，复杂多人混战，敌人突然生成，科幻感，赛博朋克感，游戏CG感，卡通感，塑料质感，夸张血腥，血浆特写，爆炸过多，镜头剧烈乱晃，快速无逻辑切镜，过曝，过饱和，白天场景，暖色阳光。

A tiny hand-drawn 2D animated explorer discovers a giant abandoned clockwork workshop, turning gears, springs, buttons, and mechanical objects into a thrilling miniature obstacle course. Style: Mixed-reality hybrid animation combining expressive hand-drawn 2D character art with a realistic mechanical workshop environment. Playful stop-motion physics, smooth hand-drawn motion, cinematic lighting, realistic textures, and detailed mechanical sound effects with energetic background music. Camera & Framing: Extreme macro close-ups, low-angle tracking shots, dynamic overhead views, shallow depth of field, tilt-shift miniature effects, and fast cinematic camera movements following the tiny explorer. Audio & Scene Breakdown [00:00] - [00:03] Visual: A tiny explorer wearing a blue jacket, brown boots, and a small red backpack stands beside a massive brass gear as it slowly begins rotating. [00:04] - [00:07] Visual: She jumps onto the moving gear teeth, runs across them, and leaps onto a giant metal spring that launches her onto a wooden workbench. [00:08] - [00:12] Visual: A huge mechanical lever suddenly moves beneath her, sending her sliding across a polished metal surface before she grabs a dangling copper wire. [00:13] - [00:17] Visual: She swings across the workshop and lands inside an enormous open pocket watch, running between its moving gears as sparks glow in the background. [00:18] - [00:22] Visual: A giant wind-up key begins turning, creating a powerful gust that pushes her across the table while loose screws and metal pieces roll around her. [00:23] - [00:27] Visual: She discovers a tiny glowing blue crystal hidden inside a mechanical box and carefully reaches toward it as the entire workshop begins coming alive. [00:28] - [00:30] Visual: She touches the crystal, causing hundreds of gears to move simultaneously as the camera rapidly pulls backward to reveal the entire workshop operating like one enormous machine. Maintain consistent character design, proportions, environment scale, lighting, textures, motion physics, and visual style throughout the entire video. Use original characters and environments only. No logos, text, watermarks, branded elements, or recognizable copyrighted characters.

日本の地域イベント。着ぐるみの頭を外す瞬間から始まる、さくらの短い休憩。乱れた前髪、汗を拭く手、水を飲んで抜ける肩の力。スタッフと目を合わせてうなずき、再び頭をかぶって会場へ。 後半は子どもたちとの交流と記念撮影。嬉しくて近づく子どもたち、並ぶまでの小さな慌ただしさ、スマートフォンを構える保護者。着ぐるみの中のさくらは、相手の高さへ手を下げ、その動きを待つ。 撮影は付き添う友人の横向きスマートフォン。軽い手ブレ、追従の遅れ、構図の迷い。演技を強調せず、行動と環境音で見せる。セリフ・字幕・BGMなし。 【着ぐるみの設計指示】 着ぐるみはオリジナルの犬型キャラクターとして設計する。既存のアニメ・ゲーム・商品・企業・自治体のキャラクターや、着ぐるみ制作会社の既存デザインを参照・模倣しない。特定キャラクターの特徴的な輪郭、耳、目・鼻・口の配置、配色、模様、衣装、装飾の組み合わせを再現しない。既存キャラの色替えや部分変更ではなく、独自の造形として構成する。

Create an ultra-realistic cinematic Japanese lifestyle vlog featuring a young Japanese woman enjoying a peaceful rainy day in a quiet Japanese neighborhood. A young woman with black hair tied in a neat ponytail, wearing a white and navy striped oversized T-shirt, beige shorts, and white sneakers, stands under a traditional covered bus stop during light rain. She looks outside, gently extends her hand to feel the raindrops, then turns around and runs playfully along a quiet residential street. Show cinematic transitions of her running through the wet neighborhood, passing traditional Japanese houses, green trees, and a peaceful park. She eventually reaches a wooden shelter in the park, adjusts her ponytail, stretches her arms, and smiles naturally toward the camera. Visual Style: Photorealistic Japanese slice-of-life cinematography, soft overcast daylight, realistic rain reflections, natural human movements, beautiful greenery, peaceful atmosphere, authentic Japanese residential environment. Camera Direction: Start with a cinematic side-profile close-up, transition into a medium shot of her touching raindrops, followed by a smooth rear tracking shot while running. Finish with emotional close-ups and a warm natural smile. Technical: 4K, 9:16 vertical, 24fps, realistic facial expressions, consistent character identity, smooth camera movement, natural physics, detailed environment, cinematic color grading. Important: Maintain the same woman's face, hairstyle, outfit, and body proportions throughout every scene. No facial distortion, no outfit changes, no unnatural movements, no text or watermark.

30-second comedy short. Two realistic house cats, a chunky orange tabby and a skinny black-and-white tuxedo cat, sit side by side on a dim rooftop ledge at night, city lights blurred behind them. Warm orange and blue lighting, shallow depth of field, mockumentary-style camera with sudden zoom-ins for comedic timing. Their mouths move naturally in sync with speech. They whisper in sad-angry, deadpan voices like two exhausted men at a bar, constantly glancing over their shoulders. Burned-in white English subtitles at the bottom of the frame, synced to speech. 0-5s: Close-up of the orange tabby, ears flat, one eye twitching. Long, dramatic sigh. Orange (whisper): "Bro... she woke me up at 5 AM to ask if I love her." Subtitle: "Bro... she woke me up at 5 AM to ask if I love her." Quick zoom-in on his tired face. 5-10s: Cut to the tuxedo, tail whipping, slowly shaking his head. Tuxedo (whisper): "Mine knocked a glass off the table, looked me dead in the eye, and blamed ME." Subtitle: "Mine knocked a glass off the table, looked me dead in the eye, and blamed ME." Orange nods slowly in the background: "Classic." 10-15s: Two-shot, both staring at the skyline. Orange (whisper): "She says I never listen. I'm a CAT. I don't listen to anybody." Subtitle: "She says I never listen. I'm a CAT. I don't listen to anybody." Tuxedo: "Respect the honesty, though." 15-20s: Tuxedo leans in, furious whisper, eyes wide. Tuxedo (whisper): "She told me to 'use my words.' I HAVE four words. All of them are 'feed me' and 'move.'" Subtitle: "She told me to 'use my words.' I HAVE four words. All of them are 'feed me' and 'move.'" Orange wheezes silently, shoulders shaking. 20-25s: Both go quiet. Orange stares into the distance. Orange (whisper): "That's it. I'm done. Tonight I'm leaving her." Tuxedo (whisper): "Finally. Say it again, louder." Orange (louder, voice cracking): "I'M LEAVING HER!" Subtitles match. 25-30s: A distant female voice calls out: "Dinnerrr!" Both cats' ears shoot up instantly. They stare at each other, then bolt off the ledge in opposite directions. Cut to the empty ledge, one tiny "meow" echoes. Orange (off-screen): "Coming, babe!" Tuxedo (off-screen): "Coming, honey!" Subtitles match. Fade to black. Style: photorealistic, 4K, cinematic color grade, subtle city ambience, light comedic sound effects (record scratch on the "four words" line, tiny whoosh on the bolt-off). No cartoon look, no text other than subtitles.

STYLE: Miniature Tiny-Person FX, photorealistic live-action, cinematic macro photography, ultra-shallow depth of field, realistic amusement-park lighting, ultra-detailed 8K, vertical 9:16, high-energy viral short-video style. DURATION: 30 seconds. SETTING: A gigantic indoor arcade filled with oversized claw machines, racing simulators, basketball games, air-hockey tables, giant plush toys, neon signs, glowing buttons, prize shelves, and colorful arcade machines. Everything feels enormous from the perspective of a palm-sized woman. CHARACTER: One athletic miniature woman (1:15 scale), consistent throughout: cropped bomber jacket, cargo pants, sneakers, compact utility backpack, same face, hairstyle, proportions, and size. She sprints across enormous glowing arcade buttons, jumping between them before launching herself from a giant joystick. She grabs a hanging cable, swings between massive arcade machines, and lands on a racing-game cabinet. She jumps onto the steering wheel of a giant racing simulator as it suddenly accelerates, holding on through the shaking motion before jumping away at the last moment. She lands on a huge air-hockey table as a massive puck races toward her. She jumps onto it, rides across the table, leaps over the goal, and lands on a pile of oversized plush toys. Inside a giant claw machine, she climbs across stuffed animals as the mechanical claw descends toward her. She jumps onto a plush bear, uses it as a trampoline, grabs the claw, and swings out of the machine. She lands on a towering prize shelf, races across giant stuffed toys, and leaps onto a glowing neon sign above the arcade entrance. The camera pulls far back to reveal her tiny scale against the enormous arcade. Large glowing text appears: “BUILT FOR THE BIG MOMENTS” She raises both arms as colorful arcade lights flash around her. SFX: Arcade beeps, button clicks, footsteps, mechanical sounds, engine roar, tire screech, air-hockey hiss, plush impacts, rushing air, electronic chimes, crowd ambience, energetic victory sound. VISUAL REQUIREMENTS: Consistent 1:15 scale, realistic physics, seamless action, photorealistic materials, authentic reflections and shadows, cinematic macro DOF, consistent character identity. No cartoon look, animation style, scale changes, duplicate character, outfit changes, distorted anatomy, floating objects, subtitles, narration, or watermark.

Create a 30-second ultra-realistic AAA open-world third-person action game sequence, fully rendered as fictional GTA-style gameplay from beginning to end. Generate the female player character yourself: a stylish young woman with long dark hair, natural facial features, dark fitted jacket, dark pants and practical boots. Keep her appearance perfectly consistent throughout the entire video. Set the mission in a highly detailed modern city bank with realistic interiors, security guards, vault, cash, police vehicles, traffic and NPCs. Use a true third-person gameplay camera behind the character, realistic game physics, animations, lighting, reflections and a complete GTA-style HUD with minimap, health, mission objectives and notifications. At 0–4 seconds, she enters the bank and the HUD shows MISSION START — BANK HEIST — STEAL THE CASH. At 4–8 seconds, she quickly bypasses security and reaches the restricted vault corridor; HUD: SECURITY BYPASSED. At 8–14 seconds, she reaches the vault, opens it and rapidly fills a duffel bag with cash; HUD: CASH SECURED. At 14–18 seconds, the alarm activates, red warning lights flash and NPCs panic as she runs toward the exit; HUD: ALARM TRIGGERED — ESCAPE. At 18–24 seconds, she exits, gets into a waiting getaway car and drives away while police vehicles begin chasing her through busy city streets. At 24–27 seconds, she makes a sharp turn through a narrow street and loses the police, with the minimap showing ESCAPE ROUTE. At 27 seconds, make exactly one hard cut to a natural third-person gameplay continuation at a hidden safehouse, where she steps out carrying the cash-filled duffel bag. HUD displays MISSION PASSED — CASH SECURED — ESCAPE SUCCESSFUL — RANK S. Make the entire video look like authentic high-budget GTA gameplay, including game-engine visuals, HUD, minimap, NPC behavior, vehicle physics and camera movement. No live-action look, no cinematic movie style, no anime, no cartoon, no random cuts, no teleportation and no graphic gore.

Create an ultra-realistic Korean zombie-horror sequence inside an old apartment elevator at night. A woman in a long golden light-cream one-piece dress enters alone, carrying a shopping bag. The elevator stops between floors, the lights flicker, and the floor indicator glitches to 13. A dark figure appears only in the mirror, whose reflection slowly becomes infected with pale skin, milky eyes, and dark veins. The mirror cracks as an infected hand reaches through. The doors open to a nonexistent dark floor where infected silhouettes rush toward her. She barely escapes as the doors slam shut. When the elevator reaches the 12th floor, she exits—but her reflection stays behind, smiles, then suddenly lunges toward the camera. Cut to black. Style: Ultra-realistic Korean zombie horror, claustrophobic elevator, cinematic handheld camera, flickering fluorescent lights, photorealistic skin, realistic reflections, tense atmosphere, dark shadows, natural facial expressions. Negative prompt: Cartoon, anime, stylized CGI, exaggerated gore, distorted anatomy, extra limbs, inconsistent face, unrealistic reflections, floating objects, text, subtitles, watermark.

ai video field test (oct 1, 2026) seedance 2.5, 1080p prompt: Create a 30-second ultra-photorealistic live-action personal vlog / home-video style film of an elderly Korean man spending a quiet ordinary summer morning in South Korea. The result should feel like a forgotten personal recording of an ordinary summer day. Not a commercial. Not a fashion film. Not a music video. Not a professional cinematic production. It should feel like someone casually captured pieces of a real morning as it happened. The emotional appeal should come from small human moments: waking up slowly, sitting on the edge of the bed, rubbing his face, putting on slippers, walking through a modest apartment, making a simple breakfast, standing quietly in the kitchen, looking out the window, and stepping outside into familiar neighborhood streets. The feeling should be quiet, masculine, nostalgic, slightly melancholic, warm, spontaneous, and deeply human. Set it in a real Korean neighborhood — modest apartment interiors, summer light, a small kitchen, quiet residential streets, low-rise buildings, local storefronts, alleyways, electric fans, sliding windows, convenience stores, crosswalk sounds, distant traffic, and subtle signs of daily life in Korea. Keep the environment grounded and unglamorous, like a normal lived-in morning. The man should feel completely real: an elderly Korean man, late 60s to 70s, ordinary appearance, gentle face, natural aging, slightly tired but calm presence, simple casual home clothes, believable posture and movement. Maintain consistent identity throughout the video. Visual style: Modern but humble handheld home-video / smartphone footage with imperfect framing, natural camera shake, slight autofocus breathing, occasional exposure shifts, and subtle observational composition. Nothing polished or overly cinematic. The camera should feel like it just happened to be there. Mood and tone: Natural, intimate, and observational. Slightly melancholic but not dramatic. Warm summer air, soft ambient light, a sense of memory and passing time. The tone should feel personal and truthful. Suggested sequence: Slowly waking up in bed in the morning heat Sitting up and gathering himself Walking to the kitchen in slippers Pouring cereal or preparing a very simple breakfast Standing quietly in the kitchen eating or drinking Looking out the window for a moment Leaving home and stepping into the neighborhood Walking slowly through familiar Korean side streets Audio: Use natural diegetic sound only — room tone, bedsheets, slippers on the floor, spoon against bowl, fridge hum, fan noise, distant traffic, birds, footsteps, quiet street ambience, a passing scooter, crosswalk beeps, neighborhood sounds. No music, no narration, no subtitles, no logos, no watermark. Important constraints: Prioritize natural behavior, consistent identity, believable physics, imperfect handheld framing, authentic environmental details, and the feeling that the camera happened to be there.

Create a cinematic, ultra-realistic 30-second live-action food sequence showing the complete preparation of crispy Korean-style glazed chicken in a warm home kitchen. Begin with an overhead close-up of raw chicken pieces being carefully cut into bite-sized chunks on a wooden cutting board. Transition smoothly to a stainless-steel bowl as the chicken is coated with a creamy white seasoned mixture, showing realistic hand movements and detailed food textures. Move into a dynamic close-up of the coated chicken being lowered into hot oil. Capture the bubbling oil, golden batter, steam, sizzling particles, and gradual transformation as the pieces become deeply golden and crispy. Use macro shots to emphasize the crunchy exterior and realistic frying physics. Cut to a dark pan as thick glossy red Korean-style sauce is poured in, followed by finely chopped garlic. Show the sauce slowly heating and bubbling, becoming rich, sticky, and glossy. Add the crispy fried chicken into the pan and toss it repeatedly with chopsticks until every piece is evenly coated in the vibrant red glaze. End with a beautiful close-up of the finished crispy chicken arranged on a white ceramic plate, covered in shiny red sauce and topped with sliced green onions and sesame seeds. Slowly push the camera toward the dish as the glaze catches the warm kitchen light, emphasizing the crunchy texture and appetizing details. Ultra-photorealistic cinematic live-action food commercial, realistic human hands, authentic Korean-style cooking, detailed food textures, crispy batter, bubbling hot oil, realistic steam and heat, glossy sauce, natural kitchen lighting, warm wooden surfaces, macro food photography, shallow depth of field, smooth camera movement, overhead shots, close-ups, realistic frying physics, rich colors, subtle film grain, appetizing presentation, physically accurate ingredients, 30 seconds, widescreen 16:9, no text, no logos, no watermark, no cartoon look, no plastic CGI appearance, no distorted hands, no artificial food textures.

Create a highly photorealistic 30-second early-2000s Tokyo street vlog filmed on an authentic consumer MiniDV/DV camcorder. IDENTITY LOCK: Use the exact same young Japanese woman from the reference image. Keep her face, hair, body proportions, outfit and dark-brown shoulder bag identical throughout. CAT LOCK: Use the SAME small fluffy domestic cat throughout. Realistic fur and movement, wearing a simple harness. The cat is always safe and unharmed. ERA LOCK: Everything must authentically belong to Tokyo around 2003–2004: streets, cars, clothing, shops, signs, vending machines and technology. No smartphones, AirPods, modern cars, LED screens or contemporary fashion. CAMERA: Genuine consumer MiniDV footage—handheld shake, imperfect framing, autofocus hunting, exposure shifts, soft digital detail, subtle DV noise, motion blur and occasional imperfect zooms. No 4K, DSLR, VHS effect, cinematic stabilization or modern retro filter. EXACT STORY: 0–5s — WALK: She casually walks down a Tokyo side street carrying her cat. 5–9s — SNATCH: A young Japanese man approaches from behind, suddenly takes the cat from her arms and runs away. She reacts in shock. The cat remains secure and unharmed. 9–15s — CHASE: She runs after him down the same street while the camcorder operator follows shakily. The man continues holding the cat safely. 15–21s — CATCH: She catches him, grabs his sleeve and firmly demands her cat back. Non-graphic confrontation, no weapons or serious violence. 21–25s — RECOVER: She takes the cat back, checks it carefully and hugs it with relief. 25–30s — WALK AWAY: She gives the man an angry look and walks away carrying her cat safely. The camera follows from behind. CONTINUITY: Same woman, cat, thief, outfit, bag and street throughout. No teleportation, duplicated people/cats, identity drift, scene skipping or reordered actions. Natural reactions and realistic movement. No blood, weapons, injury, subtitles, watermarks or text overlays. FINAL FEEL: A spontaneous early-2000s Tokyo home-video moment that unexpectedly turns into a short cat-recovery chase, captured like genuine consumer DV footage.

Create a hyper-realistic cinematic slapstick comedy in four seamless shots outside a modest two-story Korean villa on a bright afternoon, using natural handheld smartphone cinematography. Use Image Reference 1 as the woman and Image Reference 2 as the man; preserve their identities, faces, hairstyles, clothing, and proportions consistently. Shot 1: The woman stands at an open second-floor window, struggling to straighten a large white bedsheet with an unusually long bamboo laundry pole. The pole suddenly slips from her hands and falls vertically toward the narrow walkway as the camera instinctively tilts down and follows it. Shot 2: The scruffy man casually walks beneath the building, completely unaware, when the bamboo pole crashes onto his head with a loud THUNK, sending him dramatically onto the pavement. He sits up rubbing his head as the woman leans out apologetically; she says, “I’m sorry! I really didn’t mean to!” and he replies flirtatiously, “It’s okay... but you’re really beautiful.” Shot 3: The man grabs the bamboo pole, sprints toward the building, plants it like a pole vaulter, and launches himself upward toward the second-floor window. The bamboo bends dramatically under his weight as the camera cranes upward and tracks his ascent, with the man leaning forward as if attempting to kiss the woman. Shot 4: The woman recoils and shouts, “Eww... you’re so ugly! Go away!” before delivering a perfectly timed front kick into his torso. The bamboo snaps back like a giant slingshot, launching him violently into the sky as the camera whip-pans and continuously tracks his spinning body until he becomes a tiny silhouette disappearing into the clouds. Cut back to the window as the woman calmly dusts off her hands, picks up another bamboo pole, and resumes fixing the bedsheet as if nothing happened.

0:00–0:05 — Maasai village, firelit clearing: The camera moves low among a Maasai fireside gathering, weaving past men in red shúkà cloth seated in conversation, some sharing a gourd of fermented milk. It settles on a circle of clapping men as a warrior mid-leap rises impossibly high in the adumu jumping dance, beads and ochre-dyed hair swinging, firelight flaring across his face. Crash zoom to the calabash drummer's hands, striking at lightning speed. 0:05–0:10 — Whip pan to Mongolian steppe camp: A sudden fast 360-degree whip pan cuts across darkness and lands on a Mongolian steppe camp, a roaring bonfire outside a felt ger tent. A throat-singer crouches low, chest and throat resonating with a deep guttural drone, while beside him a horsehead fiddle player draws long, sorrowful notes, firelight catching the carved horse head at the instrument's neck. Rapid focus pull to the fiddle player's bow arm sawing with practiced intensity. 0:10–0:15 — Whip pan to Viking hall: A violent whip pan hurls the camera into a smoke-filled Viking longhouse, torchlight catching axe blades and shields mounted on timber walls. Bearded warriors pound wooden tankards on long tables in unison, chanting low and building, a horn-blower raising a great curved war horn and sounding a deep bellow that cuts through the hall. 0:15–0:20 — Whip pan to Plains fire circle: Another hard whip pan lands on a Native American Plains fire circle at night, a hide drum group seated tight around a large ceremonial drum, mallets striking in unified, driving rhythm. Quick whip pan to a dancer in feathered regalia spinning through rising sparks, fringe and feathers blurring with motion. 0:20–0:25 — Whip pan to Māori haka: A final sharp whip pan crashes into a Māori gathering at night, warriors performing the haka in fierce unison — stamping feet shaking loose earth, tongues bared, eyes wide, hands slapping chests and thighs in thunderous rhythm. 0:25–0:30 — Rhythmic collapse, hold on fire: The camera pulls back slightly but stays within the Māori circle, holding steady as the stamping rhythm reaches its peak, dust rising and catching firelight — the frame holds on their pounding feet and fierce expressions as the sequence closes.

Live-action 1990s sitcom look: bright, even, warm multi-camera lighting, slightly soft focus, saturated colors. 16:9. Fixed camera positions, no tracking, no dolly. One continuous scene, 26 seconds, with two hard cuts. Locations are shown in @Image1 (the house exterior), @Image2 (the renovation room) and @Image3 (the kitchen). The husband is the man from @Image4 : mid 20s, broad-shouldered, short dark brown hair, clean-shaven, composed face, clean white collared shirt with a tie, grey trousers. The wife is the woman from @Image5 : early 20s, slim, expressive face, dark brown hair tied up in a red kerchief, a splotch of paint on her cheek, denim overalls with one strap undone over a white paint-stained T-shirt. Keep both faces and outfits exactly as in their images. 0:00-0:04 (static wide shot) The house from @Image1 at dusk, calm and suburban, porch light on, lit kitchen window at frame-right. A kid on a bicycle rides past in front of the house, left to right across the frame, along the street. Dusk birdsong and the soft tick of bicycle wheels. HARD CUT. 0:04-0:12 (medium-wide shot of the room in @Image2 , then a slow pan to the right from a fixed position) The husband stands at frame-left near the stepladder and slowly looks around at the plastic on the floor, the ladder, the work table, the paint cans and the paint rollers. He nods in satisfaction. He walks toward the plastic-draped doorway at frame-right. At the doorframe he flips the light switch on the wall beside it: a click, and the room goes dark behind him. He pushes through the plastic sheeting and exits. The camera pans right with him. Sound of the switch click and rustling plastic. HARD CUT. 0:12-0:24 (one uncut take, static medium-wide shot of the kitchen in @Image3 , widening to a static long shot feel) The wife stands at the sink at frame-left with her back to the room. The husband pushes through the plastic-draped doorframe on the back-right wall and walks toward the kitchen island at center-right. He says, easy and casual: "It's coming along quickly." He keeps crossing toward the right side of the island. The wife slowly turns around to face him, arms crossed. On the kitchen island sit a paint can, a clean paintbrush, a carpenter's level, a hammer, a box of nails and a cup of coffee. 0:24-0:26 (same camera, super-fast aggressive zoom in on the wife's eyes) Her eyes grow wide and her nostrils flare. She makes a comical, exaggerated sitcom frown, brows knitted, mouth turned sharply down, indignant. Hold on her face. Audio: dusk birdsong and soft tick of bicycle wheels outside, quiet room tone inside, soft rustle of plastic, one switch click. The husband's voice is low, easy and casual. No music, no score, no background music, no singing, no laugh track. No on-screen text, no subtitles.

Create a 30-second ultra-realistic AAA open-world third-person action game mission, fully rendered as authentic GTA-style gameplay from beginning to end. Generate the female player character yourself: a stylish young woman with long dark hair, dark fitted jacket, dark pants and practical boots, keeping her appearance perfectly consistent throughout. Set the mission at a massive modern cargo ship at night, surrounded by dark ocean, shipping containers, cranes, realistic water, rain and distant city lights. Use a true third-person gameplay camera, realistic game physics, detailed game-engine graphics and a complete GTA-style HUD with minimap, mission objectives and notifications. At 0–4 seconds, the character quietly climbs onto the cargo ship and the HUD shows MISSION START — CARGO SHIP — FIND THE PACKAGE. At 4–9 seconds, she moves between stacked containers while avoiding security guards and reaches the marked container; HUD changes to CONTAINER LOCATED. At 9–14 seconds, she opens the container, retrieves a mysterious package and the HUD shows PACKAGE SECURED. At 14–18 seconds, an alarm suddenly activates, red warning lights flash and security begins searching the ship; HUD changes to ALARM TRIGGERED — ESCAPE. At 18–24 seconds, she runs through the container area toward the edge of the ship while security follows, then reaches a waiting speedboat below. At 24–27 seconds, she jumps into the speedboat and rapidly drives away across the ocean as the cargo ship recedes behind her. At 27 seconds, make exactly one hard cut to a natural third-person gameplay continuation showing her speedboat racing toward distant city lights, with the HUD displaying MISSION PASSED — PACKAGE SECURED — ESCAPE SUCCESSFUL — RANK S. Make the entire video look like high-budget GTA gameplay, including authentic HUD, minimap, NPC behavior, camera movement, water physics, vehicle physics, lighting and reflections. No live-action look, anime, cartoon, random cuts, teleportation or graphic gore.

Character consistency: Same young adult female traveler throughout, same face, hairstyle, body proportions and outfit, realistic skin, natural expressions, cinematic photorealistic travel-vlog style, smooth camera movement, realistic lighting. No face changes, no outfit changes. 0–5 sec — Arrival 🇬🇧 Young female traveler arrives in London, walking out of Heathrow Airport with a small suitcase, excited expression, cinematic establishing shot of London atmosphere, natural handheld vlog camera movement. 5–10 sec — Big Ben & Westminster She walks near Big Ben and Westminster, looks at the landmark and smiles toward the camera, quick selfie-style vlog shot followed by a wide cinematic view of Big Ben and the River Thames. 10–15 sec — Tower Bridge She explores Tower Bridge, walking across the bridge while filming herself, London skyline in the background, natural tourist atmosphere, smooth tracking shot. 15–20 sec — London Street She walks through a beautiful London street with classic red buses, black cabs and historic buildings, briefly stops at a cozy café, takes a coffee and continues exploring. 20–25 sec — London Eye She stands beside the London Eye during golden hour, looks toward the camera with a happy travel-vlog expression, camera slowly circles around her with the London skyline behind. 25–30 sec — Final London Shot She walks along the River Thames at sunset, turns toward the camera and smiles, cinematic aerial pull-back revealing London skyline, warm golden-hour lighting, beautiful travel-vlog ending. Overall: photorealistic, cinematic 4K, natural movement, realistic London environment, smooth transitions between locations, consistent character, consistent clothes, consistent face, no text, no watermark.

Main Subject: Young Korean woman, 24, naturally attractive, realistic skin, minimal makeup, long dark hair loosely down. Wearing an oversized maroon hoodie and loose casual pants, carrying a simple canvas shoulder bag. Preserve her exact identity, facial features, hairstyle, body proportions and appearance throughout. Location: Quiet older Seoul neighborhood during a pleasant morning. Narrow residential street with low-rise apartment buildings, small local shops, parked bicycles, potted plants, utility poles and modest homes. An elderly Korean woman is preparing to carry a moderately large cloth sack filled with groceries or household items. The street feels peaceful, lived-in and authentic. Lighting & Atmosphere: Warm, gentle and uplifting morning atmosphere. Soft sunlight filters between the buildings, creating warm highlights while the street remains naturally shaded. Muted colors, soft shadows, slightly faded tones and a nostalgic early-2000s feeling. Cozy and wholesome without looking staged or sentimental. Style: Ultra-realistic early-2000s Sony MiniDV home video filmed by another person. Multi-shot, but still completely candid and unstaged. Each shot should feel like the camcorder operator naturally repositioned themselves while following the moment. Natural handheld movement, subtle camera shake, imperfect framing, gentle reframing, occasional autofocus hunting, mild exposure shifts, faded colors, soft contrast, authentic DV compression, subtle digital noise and natural microphone noise. Shot 1 — 00:00–00:04: Medium-wide handheld shot from behind and slightly to the side as the young woman walks casually down the residential street. Her canvas shoulder bag rests naturally on one shoulder. She notices an elderly woman ahead struggling to lift a moderately heavy cloth sack onto her back. Shot 2 — 00:04–00:07: The camcorder moves closer from the side. The elderly woman tries to lift the sack herself, but it slips down before she can get it onto her back. She pauses and tries again. The young woman notices and naturally slows down. Shot 3 — 00:07–00:11: Closer side-angle shot. The young woman walks over and quietly gestures that she can help. She takes hold of the sack while the elderly woman steadies herself. Together, they lift it smoothly onto the woman's back. The young woman adjusts the sack slightly so it sits securely. Shot 4 — 00:11–00:14: The elderly woman gives her a grateful smile and small nod. The young woman responds with a warm little smile and a brief nod, then steps back. Shot 5 — 00:14–00:17: The young woman turns and continues walking down the street. The camcorder follows her from behind for a moment before drifting slightly to her side. Shot 6 — 00:17–00:20: She glances toward the camcorder while still walking. She gives a quiet, contented smile—natural and understated, as if she's simply pleased that she could help. She turns her gaze forward and continues walking.

Create a 30-second ultra-realistic morning vlog set in a small town neighborhood in South Korea in the early 2000s, filmed casually by a friend on an authentic consumer DV camcorder. It must feel like genuine home-video footage, not a modern video with a retro filter. Use the provided reference image for the woman's face, hairstyle and outfit. Keep her identical in every scene. EXACT ORDER — 6 SCENES: 0–5s — WASHING FACE: Early morning, soft sunrise light. The same young Korean woman (late 20s, hot pink sleeveless collared top, white high-waist jeans, white sneakers, small hoop earrings, butterfly pendant necklace, hair in a messy bun) steps out of her modest house into the small yard. An old leaking outdoor steel tap is running. She cups water in her hands, washes her face, and dries it with a small towel hanging nearby. 5–8s — GETTING READY: Fresh and smiling, she picks up a brown canvas shoulder bag and a few textbooks from a small outdoor table, glances at the camera, and heads out through the metal gate. 8–14s — THE WALK: She walks through a narrow early-2000s Korean neighborhood alley with small shops opening, an old delivery scooter, a bicycle passing. Korean elementary school kids in casual clothes, caps and yellow backpacks greet her cheerfully ("안녕하세요, 선생님!"). She smiles and waves. The same kids keep walking toward their school behind her. 14–18s — ARRIVING AT ACADEMY (HAGWON): She enters a small early-2000s private tutoring classroom with a green chalkboard, simple wooden desks, an old wall clock and a few faded posters. About 6–8 children stand up to greet her. She smiles warmly and gestures for them to sit. 18–23s — HER SHORT SPEECH: Standing in front of the class, she smiles and speaks a few warm words to the children. The children listen, smile, and clap softly. 23–30s — LECTURE BEGINS: She picks up a piece of chalk, writes the lesson title on the chalkboard, turns to the children and begins explaining. Children raise hands and take notes with pencils. She nods encouragingly. CONTINUITY: Same woman, same hot pink top, white jeans, white sneakers, hairstyle, brown canvas bag and books throughout. The same school kids appear on the road. Same children in the classroom. No teleporting, identity drift, outfit changes, disappearing props or reordered actions. ERA: Everything must look authentically early 2000s Korean: hand-painted Hangul shop signs, small neighborhood "슈퍼" grocery stores, low-rise brick villas, tangled overhead wires, old compact cars (Daewoo Matiz, Hyundai Accent, Kia Pride), classroom furniture and school supplies of that time. No smartphones, modern cars, LED screens or contemporary interiors. CAMERA: Raw consumer DV camcorder footage, 4:3 frame, with subtle handheld shake, imperfect framing, autofocus hunting, exposure shifts, soft digital detail, soft slightly washed-out skin tones, mild CCD/DV noise, motion blur, compression artifacts.

全片生物、面部近景、建筑和远景统一为明显电脑生成的高品质AAA奇幻游戏过场动画，使用数字雕刻脸模、经过美术设计的材质光影与成束头发，不混入实拍演员，不采用2D动画、赛璐璐描边或普通gameplay。整体是由连续硬切连接的末日奇幻战争蒙太奇：巨型军团推进、远超军团尺度的骨架生物显露、城内两只巨人短暂肉搏，最后切到另一只猿形巨人的咆哮蓄能；不同生物保持各自身份，不把这些切镜做成同一角色连续变形。环境与色彩：密集的中世纪半木构城市，红褐瓦顶、深棕木梁、浅米色墙面、窄石街、烟囱、尖顶塔楼与高耸外围石墙；所有普通房屋、街道和塔楼都远小于巨人，战斗者的腰胸高过附近屋顶，巨型军团更高，远处骨架生物又远超整个军团。厚重灰褐烟云和橙金逆光覆盖城市，浮尘、碎瓦和蒸汽形成多层纵深；开头为浓重焦褐尘雾，中段为暖金天空照亮的城内废墟，结尾为巨大圆形金色逆光与翻卷云雾，保持同一末日暖金色系。A巨型军团：多排无发的超巨大成年男性化人形生物，宽肩厚胸、深红肌束与浅象牙腱膜、深陷眼窝和清楚齿列，肋部及肩胸表面具有连续雕刻组织；这是生物本来的完整外观，不是新鲜伤口，没有流血。它们空手、无衣甲，前后排列有真实纵深，头肩与步幅略有差异，沿同一个方向持续缓慢行进，身体周围蒸汽升腾，脚下带起尘浪，不是静止观众或重复粘贴的同一人。B骨架巨构：军团后方仅有一个远比它们庞大的非人骨架生命体，浅褐象牙色中央脊柱高耸，两侧大量狭长弧形肋骨向外展开再向下弯曲，像覆盖军团上空的巨大骨架穹顶，肋弓之间有真实空隙，不是翅膀或建筑。前部低悬着一张狭长的红褐骨质面孔，深眼窝、尖长耳、露齿下颌，黑色长发成束垂落在面部两侧，头部通过颈部连接同一骨架，不是独立漂浮的头。最初眼睛暗，随后两眼亮起翠绿色光。C黑发硬拳巨人：中段肉搏主角之一，强健修长的人形非人生物，棕褐皮肤与清楚胸腹肌束，肩背宽而腰收，长深棕黑发、尖耳、强下颌与齿列，双眼翠绿发光。两只拳头及下段前臂覆盖浅灰白色粗糙硬化结晶，晶层依附真实拳指与腕部，仍是同一双手，不是手持石块或拳套武器；肘关节以上保留原本肌肉形态。D浅金骨甲巨人：另一名肉搏者，体量与C接近但更宽厚，短浅色头发、浅金色眼光，脸、胸、背、肩和四肢有层叠浅金褐色生物骨甲，深褐肌束连接甲片之间，甲片是身体的一部分，不是穿戴金属盔甲。空手，两臂与头部全程完整，肩颈逸出薄蒸汽。E黑鬃猿形巨人：末段独立角色，厚重黑褐毛发覆盖巨大肩背与极长双臂，躯干前倾，腿较短且强健，沉重猿人面孔、后梳深色头发、突出眉弓、尖耳与猩红眼睛，嘴中有成排尖齿。它与C、D和B是不同生物；长臂、黑毛与红眼始终保留。0.00–2.20秒：立即从浓重橙褐烟尘中的略倾斜近中景开始，前景破碎屋脊、墙板与石块遮住画面下部，数名A的巨大无发头颅与肩胸从废墟后方逼近，后面还能看见更大的模糊肋骨轮廓。摄影机沿废墟缓慢侧移，近处残墙与碎片快速横过，远处头颅较慢移动形成尺度与parallax；蒸汽和尘流不断穿过它们面前。2.20–4.20秒：切到接近地面的斜侧宽景，整排A从画面近侧延伸到远处，宽肩、长臂与肌束胸腹从烟中显出，身体随沉重步伐交替起伏，手臂自然摆动。摄影机沿军团方向缓慢退移，前景滚动尘浪，后排继续推进，不能让军团站定摆姿势。4.20–7.20秒：硬切正面低角极宽景，A的前排头肩横贯画面下方，B的巨大中央骨架与两侧肋弓在它们后方高高耸起。镜头缓慢上仰并略微前推，烟云先遮住上部结构再被气流拨开，低悬长发面孔位于中央骨架下方。两次短促金色闪电在远后方云层中亮起，瞬间照亮肋骨边缘、云团与军团轮廓后迅速衰退，闪电不是射向士兵的攻击，也不把已有生物变成其他形态。7.20–9.60秒：切B低悬面孔的正面紧近景，巨大的弧形肋骨仍在头后形成连续背景。黑发束垂在两颊旁，尖耳向外，头部微微抬正；先保留深暗眼窝，随后两眼由暗变为清晰翠绿光，视线凝向前方。烟和小尘粒持续通过镜头，不让脸冻结，不添加对白或眼部光束。9.60–11.60秒：以烟雾中的硬切转入另一片城内街区，低角正面中全景，C从灰褐尘幕中向前迈步显露，双拳低垂，胸腹与肩背随步伐运动，两侧较小塔楼和红瓦屋顶明确显示巨人尺度。镜头略向前接近，翠绿眼光透过黑发可见，两只浅灰硬化拳和前臂同时保持清晰；这不是B缩小成人形，也不复制出第二个C。11.60–12.40秒：突切向上仰拍的云天，一具蜷曲的巨大人形轮廓处于金色分叉闪电和近圆形发光蒸汽团中央，身体向下坠近，一臂抬过头侧，一膝收起。电弧沿周围空气和身体轮廓分叉，强光短暂压暗背景。只展示这个有实体身体的空中电光片段，不插入小人长大或从C体内分裂出的过程。12.40–13.20秒：切城内贴石街的倾斜低机位，远处房屋之间D已处于落地后的低蹲，石块和瓦片被震起，贴地尘浪沿街涌向镜头；落地位置与尘浪源点一致。高墙和红瓦屋顶上方可见一艘遥远的浅色长椭圆飞艇，缓慢漂移，不发射、不爆炸。镜头随地面冲击短震，屋顶和瓦片为巨人尺寸参照。13.20–14.20秒：切贴近D侧面的快速跟拍，D从低姿势向前冲出，躯干前倾、双脚交替蹬地、弯肘摆臂，肩颈蒸汽向后拖；浅金骨甲保持固定在同一身体上，双手空着，背景瓦屋和烟囱迅速后移。14.20–15.00秒：切C略低角胸上近景，C迅速转头、转肩看向冲来的D，长黑发被甩向一侧，翠绿眼睛锁定对手，下颌绷紧，身体开始转髋迎击；镜头贴近其面部短促跟转，不停下来重新对峙。15.00–16.00秒：切街道低位双人宽景，C在screen-left，D在screen-right，两者在小型房屋之间快速缩短距离。C踩稳前脚、后脚抓地，右肩和右髋向后蓄力，右侧硬化拳明显拉到身后，左臂在前方保持平衡；D继续向C压近并抬臂，瓦片和灰尘从两者脚边翻起。完整看见脚底支撑、转髋与拳头准备，不出现瞬移换边。16.00–17.40秒：摄影机快速贴过C肩侧并跟随右臂前送，C由后脚蹬地、髋部旋转带动肩和右侧硬化拳，沿单一清楚弧线打向D面侧。黑发向后甩，眼部转动带出极短翠绿拖光而不是激光；拳头逼近时短暂放慢以看清轨迹，随后清楚接触D脸侧骨甲。接触产生紧凑粉尘、小硬质碎屑与少量火星，D的头和肩先受力偏转，再带动上半身后仰。拳臂没有穿过头颅，没有血或断裂肢体，摄影机在接触瞬间短促震动。17.40–18.50秒：顺接倾斜的D近中景，他的头、肩和胸甲沿第一拳的力量方向扭转后退，两臂被甩开后开始重新收拢，脚下在瓦砾间调整支撑。镜头随上半身退移，下方红瓦屋顶保持远小于他，C仍在同一侧近处，不把这一反应剪成新变身或完全倒地。18.50–20.40秒：切回较宽的双人侧面构图，C再次把右侧硬化拳抬高至头上方形成威胁，另一只硬化拳处于较低前方。D向前顶身并...