seedance-prompting
Skill2 starsChanged 8 months agoMITView on GitHub
Generates optimized prompts for Seedance 2.0 AI video generation. Use this skill whenever the user wants to create, improve, or troubleshoot AI video prompts for Seedance or any multi-modal AI video tool. Triggers on: video prompting best practices, the @ multi-modal reference system, camera/motion replication, lip-sync or audio-driven video, multi-shot narratives, one-take sequences, beat-synced editing, video extension/editing, product/commercial video prompts, or character consistency across shots. Use even if the user just says 'video prompt' or 'AI video' without naming Seedance. Also triggers on: motion transfer, video-to-video, first-frame lock, scene transitions, choreography replication, or workflows combining images + video + audio as inputs.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- raphaelbgr/ai-video-promptingtheir own version, first seen Sep 3, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- raphaelbgr/ai-video-prompting★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 45
Every prompt benefits from this universal structure:
- Distinctive: line 104
Always append quality and stability keywords to your prompts:
- Distinctive: line 108
Stability: No blur, no ghosting, no flickering, stable footage
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Seedance 2.0 — Prompt Engineering Guide
How This Model Works
Seedance 2.0 is a multi-modal directed video generation engine. Unlike earlier text-to-video models where you type a prompt and hope, Seedance lets you direct and produce — referencing images, videos, and audio to control exactly what gets generated.
The core equation: Multi-modal @ references (reference anything) + strong creative generation + precise instruction following.
You upload assets (images, videos, audio), they get auto-labeled (@Image1, @Video1, @Audio1), and you reference them in your prompt to assign each file a specific role — character source, motion reference, camera guide, audio track, style template, first frame, etc.
Settings Reference
| Setting | Options |
| Aspect Ratio | 16:9, 9:16, 1:1, 4:3, 3:4 |
| Duration | 4s – 15s per generation |
| Reference Images | Up to 6 |
| Reference Videos | Up to 6 (total duration ≤ 15s) |
The Golden Rules
1. Assign Every Asset a Role
The @ reference system is the core of Seedance 2.0. Don't just upload files — explicitly tell the model what each file is for. Every @Image and @Video reference should have a stated purpose: character appearance, camera movement, choreography, audio source, first frame, style reference, etc.
@Image1 as the first frame, reference @Video1 for camera movement, use @Audio1 for background music.
2. Write Actions Slowly and Continuously
Seedance excels at slow, fluid, natural motion. Describe actions with adverbs that convey pace and smoothness. Avoid vague single words like "dancing" or "walking" — be specific about the motion.
Good modifiers: slowly, gently, continuously, naturally, fluidly, smoothly Good specifics: "slowly turns around," "gently raises hand," "light footsteps," "slightly lowers head," "sways with the wind" Avoid: exaggerated, high-speed, complex multi-person interactions, extreme twisting
3. Always Add Stability Constraints
Preventing face morphing and body distortion requires explicit constraint keywords in every prompt:
Character's face remains stable without deformation, normal body structure, natural and fluid motion. Clear facial features, consistent clothing, unchanged hairstyle.
4. Use the Prompt Formula
Every prompt benefits from this universal structure:
Subject + Action + Setting + Lighting + Camera Language + Style + Quality + Constraints
5. Be Explicit About Reference vs. Edit
Make clear whether you want to edit an existing video or use it as reference for something new. These are different operations and the model handles them differently.
The @ Multi-Modal Reference System
The @ system lets you reference text, images, videos, and audio as inputs. Each uploaded asset is auto-labeled and you reference it by name in your prompt. This is what gives you real creative control — you're assigning roles to every file.
For the full reference system guide with all patterns (first/last frame lock, video merging, continuous actions, audio extraction, multi-asset workflows), read references/multimodal-references.md.
Prompt Architecture by Category
Seedance supports radically different generation modes depending on what you're trying to do. Each mode has its own prompting patterns:
| Mode | When to Use | Key Pattern |
| Text-to-Video | No reference files, pure description | Detailed scene + action + camera + quality keywords |
| Image-to-Video | First frame lock, character consistency | @Image1 as the first frame + action description |
| Motion/Camera Transfer | Replicate choreography, camera work | Reference @Video1's camera movements / choreography |
| Template Replication | Copy ad format, effects, editing rhythm | Reference @Video1's rhythm and transitions |
| Video Extension | Add time to existing clip | Extend @Video1 by Xs (set duration to new portion only) |
| Video Editing | Swap characters, add elements, change narrative | Specify what changes; everything else preserved |
| Audio/Lip Sync | Dialogue, voiceover, sound effects | Natural dialogue in quotes + voice timbre references |
| Beat-Synced Editing | Music video, rhythm montage | Image sequence + sync to @Video1's rhythm and beats |
| One-Take Continuity | Unbroken shots, scene transitions | Scene sequence + one continuous take, no cuts |
| Emotion Performance | Nuanced facial/body expression | Describe emotional arc + reference expressions |
For detailed prompting patterns, example prompts, and templates for each category, read references/prompt-patterns.md.
Camera & Cinematography Language
Seedance has strong camera language recognition. Write camera directions directly in your prompts.
Shot types: Close-up, medium shot, wide shot, extreme close-up, full body, over-the-shoulder, aerial overhead, POV/first-person
Camera movements: Slow push-in, gentle pull-out, steady pan, half-orbit, full orbit, tracking shot (front/side/behind), dolly, crane, handheld-stable, whip pan, Hitchcock zoom
Stability modifiers: Fixed shot, no shake, silky smooth, stable footage, no jitter
Example: Medium shot, slow push-in, steady follow-cam, silky smooth with no jitter.
Time Segmentation for Longer Videos
For 10–15 second videos, break your prompt into time segments for precise control:
0–3s: Close-up of the character's face.
4–8s: Camera pulls back to reveal the full scene.
9–12s: Action sequence with tracking shot.
13–15s: Closing frame with brand text.This gives you shot-by-shot direction within a single generation.
Quality Control Keywords
Always append quality and stability keywords to your prompts:
Resolution: 4K, ultra-high definition, rich details, sharp resolution Visual feel: Cinematic quality, natural colors, soft lighting Stability: No blur, no ghosting, no flickering, stable footage Character: Clear facial features, stable face, no distortion, no deformation, normal body proportions
What to Avoid
- Complex multi-person interactions — fighting/running/jumping with many people at once
- Vague descriptions — "looks good," "very beautiful," "super cool" (these tell the model nothing)
- Contradictory requirements — "ultra-fast speed + extremely stable" (pick one)
- Mixing up @ references — accidentally referencing the wrong image or video number
- Single vague action words — say "slowly walks forward with light footsteps" not just "walking"
Pro Tips
- High-quality references — If @Image1 is blurry, your video will be blurry. Use 2K+ images.
- Combine Video + Image tags — Upload a photo as @Image1, a dance video as @Video1, then:
@Image1 performs the dance from @Video1.Perfect motion transfer with custom character. - Iterate small changes — Don't rewrite your whole prompt. Change one word or swap one reference.
- Label carefully — When working with many files, double-check every @ reference points to the right file. Mixing up numbers is the #1 cause of unexpected results.
- Reference audio from videos — You don't need separate audio files:
Background music references @Video1's audio. - Specify what to extract — "Reference @Video1's camera movement" hits harder than just mentioning the file. Be explicit: camera work, choreography, rhythm, effects, transitions, or voice timbre.
Reference Files
references/multimodal-references.md— The complete @ reference system: all input patterns, asset roles, merging, extending, continuous actions, and audio extractionreferences/prompt-patterns.md— Detailed prompting patterns and example prompts for every generation mode (text-to-video, motion transfer, template replication, video editing, lip sync, beat sync, one-take, emotion, multi-shot storytelling, commercial/product)references/templates.md— Ready-to-use prompt templates for common use cases (cinematic portrait, landscape, product showcase, dance transfer, narrative, extension, beat-sync, emotion)
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

