Seedance 2.0 Usage Guide: Complete Prompt Engineering Playbook
Published on February 15, 2026 by Remy
Seedance 2.0 Usage Guide: Complete Prompt Engineering Playbook
Seedance 2.0 is a multimodal AI video generation model developed by ByteDance’s Seed team, launched in February 2026. It is available through the Doubao App and Jimeng AI platforms. This guide is a hands-on playbook covering everything you need to craft effective prompts and get professional-quality video output from Seedance 2.0.
Key Capabilities at a Glance
- Multimodal input: Up to 12 reference files (9 images + 3 videos + 3 audio clips) plus text instructions
- Native audio-video sync: The only model in its class that supports audio references, with lip-sync in 8 languages
- Automatic storyboarding: Just describe your story and the model plans shots automatically
- 2K resolution, 24-60 fps, up to 15-20 seconds per generation
- 90%+ usable output rate
The Universal Prompt Formula
Every effective Seedance prompt follows this structure:
Subject + Action + Scene + Lighting + Camera Movement + Style + Quality + ConstraintsTemplate Card Format
Use this template as your starting point for any generation:
Subject: [character/object, age or material]
Camera: [shot size] + [movement type] + [angle], [focal length]
Style: [visual anchor: film stock / craft / director style], [lighting], [color treatment]
Constraints: [prohibited items], [frame rate / pacing], [duration or beat timing], [consistency requirements]Curated Prompt Examples
1. Wuxia Action Sequence
A wuxia-style male hero wearing black martial outfit, fighting enemies in a rainy
bamboo forest at night. Fast sword combos with visible sword light trails and
splashing water. Fast follow camera, crane shots, and quick close-ups. Cinematic
camera language. Maintain character appearance and clothing consistency. Realistic
physics, wet fabric, rain interaction. 4K Ultra HD, no blur, no ghosting.2. Cyberpunk Chase
A high-energy cinematic action sequence at night in a neon-lit city, the camera
tracking a lone character sprinting through rain-soaked streets as police drones
and headlights blur past. Quick cuts between close-ups of determined eyes, boots
splashing through puddles, and wide shots of traffic narrowly missing him.
Cyberpunk aesthetics, high contrast neon lighting, IMAX cinematic texture. 4K,
rich details, sharp clarity.3. Luxury Product Ad
Close-up macro shot, smooth gimbal orbit, steady motion. A matte black luxury
watch on a velvet stand rotates slowly while soft light reflects off the glass.
Style: High-end studio photography, cool-toned lighting, crisp reflections. Keep
logo sharp, no warping, tripod-stable, no background flickering. 4K Ultra HD.4. Healing Japanese Aesthetic
A young woman stands under a cherry blossom tree, gentle breeze blowing through
her long hair, she slowly looks up at the falling petals. Medium shot, slowly
pushing in to face close-up. Japanese fresh style, soft afternoon sunlight,
shallow depth of field. 4K, stable picture, no distortion.5. Film Noir
A woman in a red dress walks confidently down a rainy city street at night, neon
reflections on wet pavement, film noir style, tracking shot. Harsh chiaroscuro
lighting, deep shadows, cinematic texture, 4K. Maintain face and clothing
consistency, stable picture.6. Knight Multi-Shot Narrative
A multi-shot sequence of a knight in silver armor. Shot 1: Wide shot as he enters
a dark cave with a torch. Shot 2: Close-up of his nervous eyes. Shot 3: He draws
his sword, which glows blue. Low-angle tracking shot, 2K resolution. Sound of
crackling fire and echoing footsteps.7. Cinematic Portrait
An 18-year-old Japanese anime girl with short hair, wearing a white dress and
straw hat, standing on a forest path in warm summer afternoon sunlight. She slowly
turns toward the camera and smiles gently. A light breeze moves her hair and
dress. The camera slowly pushes in from medium shot to close-up. Soft natural
lighting, film grain, healing and peaceful mood, cinematic quality. Maintain face
and clothing consistency, no distortion, high detail.8. Traveler at Dusk (Dolly Push-In)
A weary traveler in a dusty leather cloak. He slowly removes his wide-brimmed hat
and exhales, looking toward a sunset. Low-angle medium shot, slow dolly-in towards
his face, shallow depth of field. Warm golden hour light, cinematic color grading.
4K, stable picture, no flickering.9. Coffee Ad with Timed Storyboard (Chinese)
This prompt demonstrates Seedance’s timed beat storyboarding in Chinese. It extends a reference video to 15 seconds with three timed segments: light and shadow sliding across a table (1-5s), a coffee bean falling with a dolly push-in to black (6-10s), and brand text fading in (11-15s).
将@视频1延长15秒。1-5秒：光影透过百叶窗在木桌、杯身上缓缓滑过。6-10秒：一粒
咖啡豆从画面上方轻轻飘落，镜头向咖啡豆推进至画面黑屏。11-15秒：品牌英文渐显。10. Period Drama Face-Off (Chinese Multi-Shot)
This prompt shows a multi-shot Chinese-language storyboard for a wuxia standoff scene. Shot 1: orbit around a woman in red drinking wine, then rack focus to a woman in black in the distance. Shot 2: aerial drone wide shot of the cliff. Shot 3: close-up of both characters in a tense standoff.
第一个画面：从红衣女子拿起酒壶喝酒开始，环绕运镜到红衣女子背部，然后移镜变焦
渐隐看到远处的黑衣女子。第二个画面：切换无人机航拍大全景，展现悬崖全貌。第三个
画面：两个人物的近景特写，对峙氛围紧张。电影质感，4K高清，面部清晰不变形，自然
光影。11. Full Multimodal Reference (Image + Video + Audio, Chinese)
This prompt demonstrates full multimodal referencing — combining image, video, and audio references simultaneously. It instructs the model to replicate the camera work and action from a reference video, apply it to a character from a reference image, use slow-motion focus on a thrown flying knife, and add fight/ambient sound effects with background music from an audio reference.
参考@视频1的人物动作和运镜手法，生成@图片1中黑衣人物在竹林将飞刀掷出的视频。
起始帧的视角景别严格参照@视频1，飞刀掷出以后慢动作对焦飞刀，并虚化黑衣人物。
做到1比1还原参考视频，仅人物改为黑衣女子。仅生成打斗音效和环境的音效，并加入
背景音乐@音频1。12. Product Showcase (Chinese)
This prompt shows a commercial product showcase using multiple image references. It uses @Image1 as the primary product, @Image2 for side structure reference, and @Image3 for material/texture detail, with an orbiting dolly shot.
对@图片1进行商业级摄像展示，侧面结构参考@图片2，材质细节参考@图片3，镜头环绕
推进，高级质感，4K超高清，画面稳定不变形。Camera Movement Keywords Quick Reference
| Movement Type | Keywords | Effect |
|---|---|---|
| Push In | dolly-in, push in | Camera moves toward the subject — builds tension or intimacy |
| Pull Out | dolly-out, pull out | Camera moves away from the subject — reveals environment |
| Track | track left/right | Camera moves horizontally alongside the subject |
| Pan | pan left/right | Camera rotates horizontally in place |
| Crane | crane up/down | Camera moves vertically up or down |
| Orbit | orbit, 360 orbit | Camera circles around the subject |
| Handheld | handheld | Adds subtle shake for a documentary / raw feel |
| Gimbal | gimbal | Smooth, stabilized motion for a polished look |
| Dolly Zoom | dolly zoom | Classic Hitchcock vertigo effect |
| Whip Pan | whip pan | Rapid horizontal sweep for urgency |
| Aerial | aerial sweep, drone shot | High-altitude overhead panoramic view |
Speed modifiers: Pair slow / medium / fast with a distance description, e.g. slow dolly-in, 1-2 feet.
Compound movements: Describe in timed beats, e.g. Start: slow dolly-in. Then: gentle pan right for the final 2 seconds.
Shot Size Recommendations
| Shot Size | Best For | Recommended Movement | Notes |
|---|---|---|---|
| Wide | Establishing shots, product overviews | Slow push-in or static | Avoid fast pans (causes ghosting) |
| Medium | Dialogue, UGC content | Handheld = personal feel; Gimbal = polished feel | The safest shot size overall |
| Close-up | Details, emotion | Micro push-in | Panning feels unnatural at close range; telephoto can blur the background |
Style Keywords by Genre
| Genre | Recommended Style Keywords |
|---|---|
| Daily Life / Vlog | Healing fresh, artistic, Japanese fresh, Korean mood |
| Sci-Fi / Blockbuster | Cyberpunk, Dark premium, Retro film, Dreamy soft light |
| Minimalist | Minimalist clean, Premium texture, Realistic photography |
| Cinematic | Cinematic texture, Film grain, Anamorphic lens flare |
| Film Noir | Film noir, Chiaroscuro lighting, High contrast |
| Animation / Anime | Anime style, Thick oil paint texture, Van Gogh Post-Impressionism |
Audio Prompt Keywords
| Keyword | Effect |
|---|---|
reverb | Large-space reverberation (halls, cathedrals) |
muffled | Muted sound (enclosed spaces, underwater) |
echoing | Echo effect (large chambers) |
metallic clink | Metal collision sound |
crunchy | Textured sounds like gravel footsteps |
high-pitched | High-frequency sharp sound effects |
crackling fire | Campfire crackle |
Quality Enhancement Suffix (Recommended for Every Prompt)
Append these keywords to the end of every prompt to maximize output quality:
4K, Ultra HD, Rich details, Sharp clarity, Cinematic texture, Natural colors,
Soft lighting, No blur, No ghosting, No flickering, Stable picture.Constraint Words (Essential — Prevents Deformation and Drift)
Important: Seedance 2.0 does NOT support negative prompts. You must use positive constraint statements instead.
Recommended constraint statements:
Maintain face and clothing consistency, no distortion, high detail.
Character face stable without deformation, normal human structure, natural and smooth movements.To suppress subtitle generation, add: Generate video without subtitles.
The @ Material Reference System
After uploading files, the model automatically assigns tags (@Image1, @Video1, @Audio1). Reference them explicitly in your prompt to specify their role.
| Purpose | Example Syntax |
|---|---|
| Set first frame | @Image1 as the first frame |
| Reference camera work | Reference the camera movement and transitions from @Video1 |
| Assign character roles | The woman in @Image1 as the lead; the man in @Image2 as supporting |
| Reference actions | Mimic the actions from @Video1 |
| Set background music | @Audio1 as background music |
| Extend video | Extend @Video1 by X seconds |
| Replace elements | Replace [element A] in @Video1 with [element B] |
Core Tips
- One shot, one verb — Multiple motion verbs in a single shot confuse the model.
- Limit characters to 1-2 — More than two characters causes identity confusion and consistency loss.
- Prompt length: 30-200 words — Too short lacks information; too long causes the model to ignore details.
- Always include constraint words:
Maintain face and clothing consistency, no distortion, high detail. - No negative prompts — Use positive constraint statements only.
- Append the quality suffix to every prompt.
- Start short, then extend — Generate 5-10 seconds first to confirm the result, then extend.
- Generate multiple variants — Create 2-4 variants and compare to pick the best.
- Double-check your @ tags — Multi-reference prompts are where mistakes happen most often.
- Keep reference videos short — Trim to the key segment; shorter references yield more precise results.
- Use “lens switch” for multi-shot — Creates shot transitions within a single generation.
- Prefer gentle motion words — Prioritize:
Slow,Gentle,Continuous,Natural,Smooth.
Troubleshooting Decision Tree
| Problem | Solution |
|---|---|
| Composition is wrong but action is correct | Adjust Camera (shot size + one movement) while keeping Subject and Action unchanged |
| Motion feels wrong (too shaky / too fast) | Swap handheld with gimbal (or vice versa), set explicit speed — do not change Style yet |
| Style or color drifting | Replace Style with a single stronger visual anchor, reduce adjectives |
| Recurring hand / label / lens-flare artifacts | Switch constraint words or change shot size (try medium shot instead of close-up) |
| Multiple angles are chaotic | Change “multiple camera angles” to “single tracking shot” |
Recommended Resources
GitHub Repositories
- ZeroLu/awesome-seedance — High-fidelity prompt collection (film / ads / animation / social media categories)
- YouMind-OpenLab/awesome-seedance-2-prompts — 200+ curated prompts
- dexhunter/seedance2-skill — Prompt generation Skill for Claude Code / Cursor
- hexiaochun/seedance2-api — API guide + storyboard templates
English Resources
- WaveSpeedAI Prompt Templates
- GlobalGPT: 5 Ready-to-Use Prompts
- WenHaoFree: Camera Movement Formula
- WeShop AI: Prompt Script Guide
- DataCamp: Seedance 2.0 Guide
- seedance2prompt.org — 500+ prompt library
- PromptHero: Seedance Prompts
Chinese Resources
- Zhihu: Universal Formula + @ Reference Explained
- Zhihu: Jimeng Step-by-Step Tutorial
- Zhihu: 100% Output with Claude Code
- Tencent News: Complete Operation Manual
- Bilibili: 30 Uses + Full Prompt Reveal (Bob)
- Bilibili: Beginner Guide + Prompt Templates (16k favorites)
- Bilibili: Most Complete Tutorial
- CSDN: Complete Prompt Guide