以 [@Image1] 作为首帧参考 原文：「Starting with the first frame of [@Image1].」
0–3 秒让披风男子近景喊令并开火，突出后坐力与枪管热晕 原文：「0-3 seconds: Starting with the first frame of [@Image1]. Close-up shot of the cloaked man, his expression stern and cold, shouting loudly in English through his mask: “All units, move out! Fire at will!”. He levels his rifle, the muzzle erupting in a grainy blue-white burst of flame, shell casings flying. Note the realistic recoil vibration as the stock braces against his shoulder, and the visible heat haze distortion in the air above the barrel from sustained firing.」
3–7 秒镜头拉至侧后方，让多种差异化重型载具沿 [@Image1] 对角线以不规则队形涌出 原文：「3-7 seconds: The camera quickly pulls back to a side-rear angle. The perspective spreads along the diagonal path of the armored vehicle in [@Image1], but no longer in neat formation. Amid thick smoke, a staggered cluster of heavy vehicles suddenly emerges, completely varied in model and height: some are eight-wheeled infantry fighting vehicles, some carry massive radar parabolic antennas, others are low-profile tracked tanks. All vehicles advance in an irregular, dispersed tactical formation, with each showing distinct differences in mud distribution and armor coatings.」
7–12 秒切至重装甲车内士兵开舱操作机枪，并让导弹车与坦克主炮同时开火 原文：「7-12 seconds: Quick close-up cut to a hulking heavy armored vehicle, a soldier forcefully shoving open the heavy metal hatch door with one hand, the door giving a slight rebounding shudder from its weight; the soldier agilely half-crouches inside to operate the heavy machine gun, his eyes warily scanning the surroundings. In the side-rear, a missile vehicle angles skyward, its missile igniting in an instant burst of massive white smoke and red light that blasts away the surrounding snow; the tank's main gun fires, its muzzle brake spewing black smoke and realistic fiery glow.」
12–15 秒给出高空俯瞰全景，让百余辆差异化载具在雪原上颠簸行进并拖出烟雾 原文：「12-15 seconds: Wide overhead vista. This is a chaotic yet disciplined torrent of steel, with over a hundred advanced vehicles of varied designs carving tracks of differing depths and widths across the snowy plain. The vehicles jostle with authentic bumps and sways over the uneven terrain, each trailing exhaust plumes and black smoke of varying density. The frame pulses with low-frequency engine rumbles and distant, unrelenting explosion flashes, evoking the raw oppressive weight of a battlefield.」
配置音效层并写入禁止项约束 原文：「Sound Design: Authentic booming English commands, thunderous tank cannon roars, the shrieking hiss of missile ignition, the cracking grind of tracks crushing hard ice, layered with epic war-drum orchestration. Prohibited: Watermarks, subtitles, uniform square formations, repeated models.」 未给项：工具平台=原帖未给；模型版本=原帖未给；时长=0–15秒；画幅=2.35:1；分辨率=4K；参考素材=[@Image1]
在 OpenArt AI 平台选用 Seedance 2.0 原文：「Seedance 2.0 on OpenArt AI」
上传参考图片并在提示词中将其引用为 @Image 1 原文：「from the reference image @Image 1」
在提示词中添加角色一致性约束，要求主体始终与 @Image 1 完全一致 原文：「Character Consistency: The subject is the young male influencer from the reference image @Image 1. In every time period and from every angle, the face shape, facial features, and skin tone must match @Image 1 exactly. Do not alter or distort the face.」
将画幅设置为 9:16 竖屏 原文：「Format: 9:16 vertical.」
在提示词中写入着装要求：每场景穿黑色时尚休闲/街头/西装造型，且每段更换服装发型 原文：「Outfit: In every scene, wear stylish and luxurious black casual, streetwear, or suit outfits. Do not wear the outfit from the character reference sheet. Completely change the outfit and hairstyle for each segment.」
在提示词中写入拍摄与画面风格约束：快节奏剪辑、手持 iPhone 竖屏质感、自然抖动与变焦、保留真实皮肤纹理、禁用美颜与电影调色、不添加屏幕文字 原文：「Camera / Style: Fast editing rhythm with cuts every 0.5 to 1 second. iPhone handheld vertical shooting texture. Mix digital zoom in, zoom out, and tilt up, with natural camera shake. Include autofocus hunting, indoor and outdoor lighting exposure changes, and image quality degradation during zoom adjustments. Preserve real skin texture including pores, beard shadow, flyaway hairs, and natural skin oil. Do not use beauty filters, excessive skin retouching, CGI textures, or cinematic color grading. Do not merge cuts or omit scenes. Do not insert on screen text.」
在提示词中写入声音要求：潮流嘻哈或 R&B 背景音乐混合环境音 原文：「Sound: Trendy hip hop or R&B background music mixed with natural environmental sounds such as street noise, footsteps, and everyday ambient sounds.」
按时间顺序生成 M-01 段落，依次执行 0–15 秒六个场景镜头指令并以指定定帧结束 原文：「M-01. Hongdae Busking Street & Tattoo Shop Tour 0 to 2 seconds: [Hongdae Walking Street] A low angle shot holding a selfie stick high above while quickly spinning to capture both yourself and the crowd in the middle of a busy busking street. 2 to 4 seconds: [Tattoo Shop Waiting Room] A tight shot pointing at the tattoo designs hanging on the wall with your finger, followed by a quick transition to your face looking at the camera while lightly biting your lip with a slightly nervous expression. 4 to 6 seconds: [Tattoo Workstation] A forearm shot rolling up the sleeve of an oversized black T shirt, rapidly cross edited with a shot of your face smiling while enduring the pain. 6 to 8 seconds: [In Front of the Shop Mirror] A tight shot checking the linework of the newly completed tattoo in the mirror, followed by your face making eye contact with the camera through the mirror and nodding with satisfaction. 8 to 10 seconds: [Hongdae Alley] A close up taking one bite of a street waffle, then blowing on it because it is hot while smiling at the camera. 10 to 12 seconds: [Select Shop Hallway] A handheld tracking shot from behind as you walk slowly with a shopping bag over one shoulder. The screen naturally shakes in sync with your footsteps. 12 to 15 seconds: [Hongdae Station Entrance] In front of the station exit, playfully extend your fist toward the camera in a fist bump pose. End with a front facing full body shot frozen on the final frame.」 未给项：工具平台=OpenArt AI；模型版本=Seedance 2.0；时长=15秒；画幅=9:16 vertical；分辨率=原帖未给；参考素材=@Image 1
在 Higgsfield 中使用 Seedance 2.0 来运行这些提示词。 原文：「Test out these prompts with Seedance 2.0 in Higgsfield: https://higgsfield.ai/?fpr=atomic」
若提示词需要，上传自己的视频、图片或音频等媒体素材。 原文：「Most of these prompts will use media upload, like video, images and audio, so make sure to add your own media if the prompt requires it.」
输入跨媒体运动迁移提示词，将视频中的动作套用到图片角色上。 原文：「Use the motion from the video in @video_1 with the characters in @image_1」
编写包含标题、基调、风格、画幅与时长，并按时间轴分镜的产品广告提示词。 原文：「TITLE: Eclipse Headphones TONE: Premium minimalism, elegant tension STYLE: Photoreal macro product cinematography FORMAT: 16:9 DURATION: 8 seconds [0–2 s | HOOK] Macro 85 mm shot gliding over brushed-metal ear-cup. Soft edge light wraps around logo. Mist of particles drifts. Ambient hum + low bass swell. [2–4 s | SETUP] 35 mm side tracking — model’s silhouette lifts headphones in slow motion. Light gradient fades from blue → gold. Sub-bass pulse builds. [4–6 s | ACTION] Tight 50 mm close-up as ear-cup seals over ear. Sound drops to silence. Camera pushes in microscopically, reflection catching logo flare. [6–8 s | RESOLUTION] Wide 24 mm reveal — subject lost in soundscape; particles dissolve into brand emblem. Subtle logo sting + fade. Continuity: Soft diffused key light, desaturated palette with gold accents, macro focus transitions.」
用图片占位符分别指定品牌代言人、产品主图与生活方式氛围，生成 15 秒广告分镜提示词。 原文：「<<<image_1>>> as brand spokesperson. <<<image_3>>> as product hero shot. <<<image_2>>> for lifestyle aesthetic. Generate 15s spot: Shot 1: Lifestyle moment with <<<image_3>>> in <<<image_2>>> warm setting. Shot 2: Problem presentation - <<<image_1>>> concern. Shot 3: Product reveal from <<<image_3>>> perfect angle. Shot 4: <<<image_1>>> transformation using product. Shot 5: Wide shot <<<image_1>>> satisfied conclusion, logo space. Commercial structure, <<<image_1>>> relatable, <<<image_3>>> pristine, <<<image_1>>> / <<<image_3>>> combined.」
按时间戳分段编写场景、镜头与对白，生成带时序的动画短片提示词。 原文：「2d cartoon style, [00:00–00:02] Wide shot — ancient temple interior glowing with golden light, walls etched with runes. [00:02–00:04] The explorer proudly declares: “Don’t worry, I studied the traps!” [00:04–00:06] Close-up — detective crosses her arms: “You also triggered them last time.” [00:06–00:08] The explorer steps on a tile — click! Arrows launch from the walls. [00:08–00:10] Slow-motion — both duck, hats flying, arrows zip past dramatically. [00:10–00:12] End shot — they crawl out covered in soot. The detective sighs: “Next time, I lead.”」
编写轨道旋转提示词，设置约 15 秒时长、顺时针 360° 匀速环绕，并分年龄段描述角色变化。 原文：「Style: Cinematic realism Orbit Rule: Slow, steady clockwise 360° Length: ~15 seconds 🎥 CAMERA INSTRUCTION Camera begins orbiting clockwise around a single character standing in open field at golden hour. Orbit speed constant. No cuts. Character ages progressively across full rotation.」
指定 Image 1 为起始帧、Image 2 为结束帧，生成 12 秒连续 FPV 镜头提示词。 原文：「Use Image 1 <<<image_1>>> as the exact starting frame and Image 2 <<<image_2>>> as the exact final frame. Do not change either image. The video must start matching Image 1 and end matching Image 2 perfectly. Generate a continuous FPV camera shot lasting 12 seconds.」 未给项：工具平台=Higgsfield；模型版本=Seedance 2.0；时长=8 seconds / 15s / ~15 seconds / 12 seconds；画幅=16:9；分辨率=原帖未给；参考素材=video、images、audio（如 @video_1、@image_1、Image 1 / Image 2 等）
在提示词中加入 POV 段落，设定第一人称厨师视角与画面构图 原文：「POV: First-person chef perspective. Only show cute female hands and arms (no face, head, or body). Delicate hands with pastel pink nails and frilly chef sleeves. Pink checkered countertop, cinematic camera movement, macro food shots, shallow depth of field.」
在提示词中写入 RECIPE 与 FOOD 段落，给出食谱步骤并强调食物质感 原文：「Follow the reference omelet recipe: Crack 3 eggs, add 2 tbsp milk, salt and pepper, whisk fluffy. Melt butter in pan, pour eggs, add cheese, mushrooms, red tomatoes, and green herbs. Fold into a golden fluffy omelet and garnish with herbs. FOOD: Make it extremely delicious: soft airy egg texture, melted cheese stretch, glossy mushrooms, juicy tomatoes, fresh herbs, gentle steam, satisfying cooking sounds.」
在提示词中写入 BUNNY HELPER 与 SEQUENCE 段落，设定兔子助手动作与 0–15 秒时间线 原文：「Add a tiny fluffy white anime bunny chef wearing a mini chef hat and apron. Cute, clumsy, playful helper: tries carrying ingredients, drops a mushroom then fixes it, struggles with a spoon, helps sprinkle herbs, gives a tiny thumbs-up, happily hops beside the finished omelet. Cute mascot style, not realistic. SEQUENCE: 0-5s: Female hands start cooking, crack eggs, bunny joins excitedly. 5-10s: Whisk eggs, butter melts, omelet cooks with fillings and close-up sizzling shots. 10-15s: Hands fold omelet, bunny celebrates, final hero shot of beautiful fluffy omelet with cozy sparkles.」
在提示词中加入 QUALITY、MOOD 与 NEGATIVE 段落，规定画质参数、氛围与排除项 原文：「QUALITY: 4K UHD 3840x2160, 24fps cinematic animation, smooth hand movement, premium anime rendering, crisp textures, beautiful lighting, cinematic depth of field. MOOD: Fluffy • Savory • Cute • Playful • Heartwarming • Cozy. NEGATIVE: No face, no full body, no realistic human, no scary bunny, no dark mood, no burnt food, no wrong ingredients, no extra fingers, no deformed hands, no distorted bunny, no text, no watermark, no blur.」 未给项：工具平台=@capcutapp；模型版本=Seedance 2.0；时长=15-second；画幅=原帖未给；分辨率=4K UHD 3840x2160；参考素材=the provided cute omelet recipe image