The woman in this clip is the size of a palm. She's free-running through a digital-devices store — vaulting off a keycap bag, parkouring across a bottle of screen cleaner, zip-lining over headphone boxes on a shopping-basket cord, surfing a rolling can of compressed air, and finally sticking a fist-raised landing on top of a glass display case while gold Chinese characters flash "挑战完成" (Challenge Complete) behind her. Five shots, one 15-second Seedance 2.0 generation, and the same 1:15-scale character in every single frame. This post is the original prompt that produced it, broken down by what each part is actually doing.
Multi-shot consistency is the part most "miniature person" prompts never solve. Ask for a tiny character doing several things and you typically get a different face, a different outfit, or a different scale every time the camera cuts. This prompt avoids that by writing the whole 15 seconds as a single shot list — one character sheet, then five numbered beats, each with its own camera move and its own sound design — instead of one vague scene description and hoping the model holds continuity on its own.
The full prompt
风格：微型小人特效，逼真的实景拍摄，电影般的微距摄影，超浅景深，明亮的现代数码设备商店，逼真的荧光照明，超详细8K，竖屏9:16，动态病毒式短视频风格。
时长：15秒。
场景：一家巨大的现代数码设备商店，塞满了高耸的货架，上面摆放着色彩缤纷的键帽包装袋、相机盒、镜头盒、耳机盒、键盘、鼠标、屏幕清洁液、清洁气罐、存储卡、显示器、购物篮和悬挂的价格标签。从一个手掌大小的女性的视角来看，一切都显得巨大无比。
角色：一位运动型的微型女性（1:15比例），整个过程中外貌一致：轻便冒险夹克、工装裤、运动鞋、多功能背包，脸部、发型、比例和尺寸在每个镜头中都相同。
00:00–00:03 — 数码货架荡秋千
她站在一个巨大的彩色键帽包装袋上，冲刺并跃向一个悬挂的价格标签，用双手抓住它，荡过货架通道，然后发射到下一个货架上。标签在她身后摆动。
镜头：低角度微距跟踪。
音效：塑料皱褶声、小脚脚步声、呼啸的风声、摆动的标签声。
00:03–00:06 — 瓶盖跑酷
她降落在一个巨大的屏幕清洁液瓶上，并在色彩缤纷的瓶盖上进行跑酷。一个瓶子在她身下摇晃；她重新获得平衡，并向前翻滚到下一个货架上。
镜头：快速侧向跟踪微距。
音效：脚步声、瓶子吱吱声、液体晃动声、硬着陆声。
00:06–00:09 — 购物篮滑索
她抓住一根悬挂的促销绳索，在相机盒、耳机盒和键帽包装袋上方滑索。恰到好处地松手，她降落在一个移动的购物篮内，周围是巨大的耳机盒、存储卡包装和镜头清洁套装。
镜头：从头顶到低角度的过渡。
音效：滑索哨声、风声、篮子轮子声、撞击声。
00:09–00:12 — 失控清洁气罐
一个巨大的金属清洁气罐在篮子内滚动。她跳到上面，像冲浪者一样保持平衡，并骑着它穿过光滑的地板，险险避开购物者的鞋子、篮子和散落的数码配件包装。
镜头：超低地板水平微距跟踪。
音效：金属滚动声、风声、商店环境音。
00:12–00:15 — 胜利终章
利用滚动的清洁气罐作为发射坡道，她跃向玻璃展示柜把手，荡上去，并得意地站在展示柜顶部。巨大的数码设备商店在她身后延伸，同时出现闪耀的粒子，大号金色中文文字显示：
"挑战完成"
她举起一只拳头，同时镜头缓缓拉远，强调她的微小比例。
镜头：电影式拉远。
音效：跃起呼啸声、金属抓握声、胜利钟声、闪耀声、商店环境音。
视觉要求：保持一致的1:15微型比例，逼真的物理效果，流畅连续的动作，逼真的包装，真实的反射，电影式微距景深，以及一致的角色身份。无动画、卡通、旁白、字幕、服装变化或比例不一致。
It's written in Chinese and Seedance 2.0 renders it without any translation step — the model doesn't care what language the prompt is in. Here's what each block is doing, in order.
1. Lock the character before the first shot
角色：一位运动型的微型女性（1:15比例），整个过程中外貌一致：轻便冒险夹克、工装裤、运动鞋、多功能背包，脸部、发型、比例和尺寸在每个镜头中都相同。
("Character: an athletic miniature woman at 1:15 scale, consistent appearance throughout — lightweight adventure jacket, cargo pants, sneakers, utility backpack, same face, hairstyle, proportions and size in every shot.")
This paragraph exists before any scene description, and it's doing one job: giving the model a single character sheet to reuse for the next fifteen seconds instead of re-imagining her for each beat. The scale ratio is stated as a number ("1:15"), not a vague "tiny" or "miniature," and the outfit is specific enough that there's nothing left for the model to improvise between shots. The closing visual-requirements line repeats the same constraint from the other direction — "consistent character identity... no costume changes or scale inconsistency" — so the rule gets stated at both the top and the bottom of the prompt.
2. Build a world out of oversized everyday objects
场景：一家巨大的现代数码设备商店，塞满了高耸的货架，上面摆放着色彩缤纷的键帽包装袋、相机盒、镜头盒、耳机盒、键盘、鼠标、屏幕清洁液、清洁气罐、存储卡、显示器、购物篮和悬挂的价格标签。
A digital-devices store is a good set precisely because nothing in it is exotic — keycap bags, screen-cleaner bottles, cans of compressed air, a shopping basket. At 1:15 scale, ordinary retail clutter turns into a parkour gym: a price tag becomes a rope swing, a bottle cap becomes a balance beam, a rolling can becomes a surfboard. The "giant everyday object" trick is doing the visual work a more exotic set couldn't — the humor and scale both come from the audience recognizing what the object actually is.
3. Write the five shots like a stunt call sheet
Rather than one continuous camera direction, the prompt breaks the 15 seconds into five labeled beats, each timestamped and each carrying its own camera move and its own sound design:
| Beat | Action | Camera | Audio |
|---|---|---|---|
| 00:00–00:03 | Swing from a hanging price tag between shelves | Low-angle macro tracking | Plastic crinkle, small footsteps, whistling wind, swinging tag |
| 00:03–00:06 | Parkour across wobbling bottle caps | Fast lateral macro tracking | Footsteps, squeaking bottle, liquid sloshing, hard landing |
| 00:06–00:09 | Zip-line over boxes into a moving shopping basket | Overhead-to-low-angle transition | Zip-line whistle, wind, basket wheels, impact |
| 00:09–00:12 | Surf a rolling can of compressed air across the floor | Ultra-low floor-level macro tracking | Metal rolling, wind, store ambience |
| 00:12–00:15 | Launch off the can onto a display case for the finale | Cinematic pull-back | Launch whoosh, metal grab, victory chime, sparkle, store ambience |
Each beat hands off physically into the next one — she lands from the swing into the bottle-cap run, rolls off the bottle run into the zip-line, drops from the zip-line onto the rolling can, launches off the can into the finale. That handoff is why five separate shots read as one continuous chase instead of five disconnected clips stitched together.
4. Script the finale as a payoff, not just another beat
利用滚动的清洁气罐作为发射坡道，她跃向玻璃展示柜把手，荡上去，并得意地站在展示柜顶部... 同时出现闪耀的粒子，大号金色中文文字显示："挑战完成"
The first four beats are momentum and near-misses; only the fifth one resolves. Giving the ending its own payoff — a stuck landing, a pull-back camera move, sparkle particles, and an on-screen text reward — is what turns a parkour run into a finished story instead of a loop that could have stopped at any beat. It's the same principle as scripting an explicit destination in a one-shot flythrough (see our breakdown of a one-shot Seedance 2.0 POV flight): a multi-beat sequence needs an ending written into the prompt, or it just stops.
5. Close with a hard consistency clause
视觉要求：保持一致的1:15微型比例，逼真的物理效果，流畅连续的动作，逼真的包装，真实的反射，电影式微距景深，以及一致的角色身份。无动画、卡通、旁白、字幕、服装变化或比例不一致。
The last line is a checklist, not flavor text: consistent 1:15 scale, realistic physics, smooth continuous motion, realistic packaging, real reflections, cinematic macro depth of field, consistent identity — then a short ban list for the failure modes that actually show up in miniature-character prompts (animation/cartoon rendering, narration, burned-in subtitles, costume drift, scale drift). Every item on that list maps back to something stated earlier in the prompt; this section just makes it non-negotiable.
Run it on hiapi
One async task call, then poll for the result:
# 1. Create the task
curl -X POST https://api.hiapi.ai/v1/tasks \
-H "Authorization: Bearer $HIAPI_TOKEN" \
-H "Content-Type: application/json" \
-d '{
"model": "seedance-2.0",
"input": {
"prompt": "风格：微型小人特效，逼真的实景拍摄... (the full prompt above)",
"aspect_ratio": "9:16",
"duration": 15,
"resolution": "1080p",
"generate_audio": true
}
}'
# 2. Poll until status = success, then download output[0].url
curl https://api.hiapi.ai/v1/tasks/$TASK_ID \
-H "Authorization: Bearer $HIAPI_TOKEN"
Two practical notes:
- Cost scales per second, by resolution tier. As of October 2026: $0.136/s at 480p, $0.293/s at 720p, $0.729/s at 1080p, $1.486/s at 4k. A 15-second clip like this one runs about $2.04 at 480p, $4.40 at 720p, $10.94 at 1080p, or $22.29 at 4k — same prompt, pick the tier that matches what the shot is for. Current rates are on the pricing page and the Seedance 2.0 model page.
- Vertical works the same as horizontal. This clip renders at
aspect_ratio: "9:16"instead of the usual16:9— the per-second pricing and duration limits don't change with orientation, only the frame shape does. For the full parameter table (first/last frames, reference images, other aspect ratios), see the Seedance 2.0 API guide.
FAQ
Can Seedance 2.0 really keep one character consistent across five different stunts in a single generation?
Yes, within a single generation — which is exactly what this prompt relies on. The character sheet ("1:15 scale, same face, hairstyle, proportions and size in every shot") comes before any of the five beats, and the visual-requirements line repeats the same constraint at the end. Drop either one and scale/identity drift between shots is the most common failure.
Does the "挑战完成" text overlay come from the same generation?
Yes — it's written into the finale beat as a scene element ("大号金色中文文字显示：'挑战完成'"), the same way the sparkle particles and victory chime are. Nothing in the clip above was composited in afterward.
Is 9:16 vertical available at every resolution tier?
Seedance 2.0 supports vertical and horizontal aspect ratios at the same resolution tiers; pricing is per-second by resolution, not by orientation. Check the model page for the current aspect-ratio list before you lock a shot into a specific frame shape.
What if I want more than five beats, or longer than 15 seconds?
15 seconds is this model's maximum single-generation duration. For a longer sequence, either plan it as multiple 15-second generations with matching character-sheet text in each prompt, or see Seedance 2.5 vs Seedance 2.0 for what changes if you need a longer single shot.
The structure here is reusable for any "giant world, tiny character" idea: a character sheet with a stated scale ratio, a set built from objects your audience already recognizes, a numbered shot list where each beat hands off into the next, a scripted finale with its own payoff, and a closing consistency clause that repeats the rules you can't afford to lose. Swap the store for a kitchen, a garden, or a desk, keep the five pieces intact, and run it against Seedance 2.0 on hiapi.
