seedance-characters

Lock character identity, assign @Tag references, and maintain face and hand consistency across multi-character scenes in Seedance 2.0. Covers 360-degree consistency testing and first-frame art direction for image-to-video. Use when a character changes appearance between shots, when building multi-person scenes, or when hands or faces are distorting.

492 stars

Best use case

seedance-characters is best used when you need a repeatable AI agent workflow instead of a one-off prompt.

Lock character identity, assign @Tag references, and maintain face and hand consistency across multi-character scenes in Seedance 2.0. Covers 360-degree consistency testing and first-frame art direction for image-to-video. Use when a character changes appearance between shots, when building multi-person scenes, or when hands or faces are distorting.

Teams using seedance-characters should expect a more consistent output, faster repeated execution, less prompt rewriting.

When to use this skill

  • You want a reusable workflow that can be run more than once with consistent structure.

When not to use this skill

  • You only need a quick one-off answer and do not need a reusable workflow.
  • You cannot install or maintain the underlying files, dependencies, or repository context.

Installation

Claude Code / Cursor / Codex

$curl -o ~/.claude/skills/seedance-characters/SKILL.md --create-dirs "https://raw.githubusercontent.com/Emily2040/seedance-2.0/main/skills/seedance-characters/SKILL.md"

Manual Installation

  1. Download SKILL.md from GitHub
  2. Place it in .claude/skills/seedance-characters/SKILL.md inside your project
  3. Restart your AI agent — it will auto-discover the skill

How seedance-characters Compares

Feature / Agentseedance-charactersStandard Approach
Platform SupportNot specifiedLimited / Varies
Context Awareness High Baseline
Installation ComplexityUnknownN/A

Frequently Asked Questions

What does this skill do?

Lock character identity, assign @Tag references, and maintain face and hand consistency across multi-character scenes in Seedance 2.0. Covers 360-degree consistency testing and first-frame art direction for image-to-video. Use when a character changes appearance between shots, when building multi-person scenes, or when hands or faces are distorting.

Where can I find the source code?

You can find the source code on GitHub using the link provided at the top of the page.

SKILL.md Source

# seedance-characters

Character fidelity, identity anchoring, and first-frame art direction for Seedance 2.0.

## Scope

- Reusable character card format
- Identity anchoring via @Tag
- Multi-character separate reference pattern
- Prop/weapon separation from character body
- Hand and face safety
- 360° consistency testing
- First-frame composition rules for I2V

## Out of scope

- Style and visual aesthetic — see [skill:seedance-style]
- Camera positioning — see [skill:seedance-camera]
- Fight choreography — see [skill:seedance-motion]

---

## Character Card Format

Write once. Reuse across all prompts for this character. Never change nouns mid-project.

```
[Name]: [age range], [build], [skin tone], [hair style/color], 
[defining features], [wardrobe description], [emotional energy].

Example:
Maya: woman mid-30s, lean build, warm brown skin, short natural hair,
sharp eyes, leather jacket over white tank, calm and precise energy.
```

---

## Identity Anchoring

For I2V and R2V, always assign the character reference explicitly:

```
@Image1's character as the subject
@Image1 for facial features and clothing
Use @Image1 and @Image2 for the character's appearance from multiple angles
```

A bare `@Image1` with no role instruction is weak.

---

## Multi-Character Patterns

For two characters, use separate identity anchors:

```
Character A references @Image1.
Character B references @Image2.

Character A throws a right punch at Character B.
Character B blocks with crossed arms.
```

Attribute every action by name. Never use ambiguous pronouns in multi-character prompts.

---

## Prop and Weapon Separation

Upload character body and prop/weapon as separate references:

```
Character appearance references @Image1.
Weapon design references @Image2.
```

This prevents the model from blending weapon details into the character's body geometry.

---

## Hand Safety

If hands are not essential to the action: frame waist-up or specify `"hands not in frame."`

If hands are essential: specify one simple action only.

```
✅  picks up the glass with right hand
✅  places hand flat on the table
✅  open palm facing camera
❌  intricate finger gestures
❌  typing on keyboard (close-up)
```

---

## Face Stability

- Prefer medium close-up with steady, locked camera for dialogue
- Avoid rapid head turns combined with extreme close-up
- Re-upload the original face reference when extending clips
- Never rely solely on the last frame of a previous clip to maintain face identity

---

## 360° Consistency Test

Before committing to a character reference, generate the same character from multiple angles (front, side, three-quarter, back). Place results side by side.

If identity holds across all angles → the reference is production-ready.
If identity drifts at any angle → improve the reference or generate from a better image.

---

## First-Frame Art Direction (I2V)

The first frame is the primary creative contract for I2V. Everything follows from it.

### Composition rules for I2V first frames

1. **Bake the camera angle.** If you want low angle, compose the first frame from low angle. Do not contradict it in the prompt.
2. **Bake the lighting direction.** The model maintains established lighting. If you want side-lighting, the first frame must show it.
3. **Pose at the START of motion.** If the character swings a sword, pose them at wind-up, not mid-swing.
4. **Clean, depth-separated background.** Cluttered backgrounds warp during camera moves.
5. **Match aspect ratio.** Generate the first-frame image in the same AR as the target video.

### What goes in the image vs. the prompt

| In the first-frame image | In the prompt |
|--------------------------|---------------|
| Character identity + costume | Motion (what changes) |
| Pose at start of action | Timing (when things happen) |
| Camera angle + lighting | Camera movement (how frame evolves) |
| Environment composition | Sound |
| Color palette | Constraints |

### Common first-frame mistakes

```
❌  Wrong lighting direction → forces re-light → causes flicker
❌  Character mid-action → no room for motion in prompt
❌  Complex cluttered background → warp during camera movement
❌  Low resolution → model loses detail for consistency
```

---

## Real-Face Restriction

Real human face references require identity verification on the Dreamina platform. Use AI-generated portraits or illustrated/3D character references instead. See [skill:seedance-prompt] for content policy.

---

## Agent gotchas

1. If identity drifts mid-clip: add `"@Image1's character as the subject"` and reduce motion complexity.
2. Character card nouns are contractual. Renaming "wool coat" to "heavy jacket" mid-project breaks consistency.
3. One hero subject per shot. Two max if interaction is essential.
4. When extending a clip, always re-upload the face reference image. The last frame is not enough.
5. The 360° consistency test is cheap insurance. Run it before committing to a production pipeline.

Related Skills

seedance-vocab-zh

492
from Emily2040/seedance-2.0

Look up 550+ Chinese cinematic, photography, and production terms with English equivalents and prompt-ready phrases for Seedance 2.0 across 22 categories, including filter-safe vocabulary for action, weapons, and combat. Use when writing prompts in Mandarin, translating a production concept into Chinese Seedance vocabulary, or verifying term accuracy.

seedance-vocab-ru

492
from Emily2040/seedance-2.0

Look up 450+ Russian cinematic, photography, and production terms rooted in the Eisenstein/Tarkovsky Soviet montage tradition, with English equivalents and prompt-ready phrases for Seedance 2.0 across 20 categories, including filter-safe vocabulary for action, weapons, and combat. Use when writing prompts in Russian or translating production concepts into Russian Seedance vocabulary.

seedance-vocab-ko

492
from Emily2040/seedance-2.0

Look up 450+ Korean cinematic, photography, and production terms with English equivalents and prompt-ready phrases for Seedance 2.0 across 20 categories, including filter-safe vocabulary for action, weapons, and combat. Use when writing prompts in Korean or translating production concepts into Korean Seedance vocabulary.

seedance-vocab-ja

492
from Emily2040/seedance-2.0

Look up 450+ Japanese cinematic, photography, and production terms with English equivalents and prompt-ready phrases for Seedance 2.0 across 20 categories, including filter-safe vocabulary for action, weapons, and combat. Use when writing prompts in Japanese or translating production concepts into Japanese Seedance vocabulary.

seedance-vocab-es

492
from Emily2040/seedance-2.0

Look up 450+ Spanish cinematic, photography, and production terms — Castilian and Latin American — with English equivalents and prompt-ready phrases for Seedance 2.0 across 20 categories, including filter-safe vocabulary for action, weapons, and combat. Use when writing prompts in Spanish or translating production concepts into Spanish Seedance vocabulary.

seedance-vfx

492
from Emily2040/seedance-2.0

Specify VFX physics contracts, energy effects, particle systems, destruction physics, and multi-layer VFX hierarchies for Seedance 2.0. Use when adding explosions, fire, water, lightning, magic effects, or any physically simulated element to a scene.

seedance-troubleshoot

492
from Emily2040/seedance-2.0

Diagnose and fix failing or low-quality Seedance 2.0 prompts using a 5-step diagnostic tree. Use when a prompt is consistently failing, producing generic output, or being rejected. Covers blurry/jittery output, camera chaos, character drift, stiff action, and ambiguous results.

seedance-style

492
from Emily2040/seedance-2.0

Control visual style, render-engine tokens, animation registers, period aesthetics, CGI material contracts, and style transfer via reference for Seedance 2.0. Use when setting a specific look — cinematic, anime, 3D, vintage, photorealistic — or when style is inconsistent across a shot chain.

seedance-recipes

492
from Emily2040/seedance-2.0

Apply genre recipe templates to Seedance 2.0 — product ads, fight scenes, brand films, mood pieces, dialogue clips, one-take journeys, music videos, novel adaptations, architecture walkthroughs, and action transfers. Use when you need a ready-made prompt structure for a known genre or format.

seedance-prompt

492
from Emily2040/seedance-2.0

Construct a production-ready prompt for Seedance 2.0 using the Director\'s Formula. Use when a user has a clear vision and needs to translate it into a genre-aware, intent-driven prompt. Covers genre routing, I2V gate, 30-100 word target, physics language, and anti-slop check.

seedance-prompt-short

492
from Emily2040/seedance-2.0

Build, validate, and compress Seedance 2.0 prompts to a hard 2000-character limit using the Five-Layer Stack, @Tag delegation, and the new Compression Engine. Use when constructing or debugging any T2V, I2V, V2V, or R2V prompt for the short-form workflow.

seedance-pipeline

492
from Emily2040/seedance-2.0

Integrate Seedance 2.0 with ComfyUI nodes and post-processing chains covering upscale, frame interpolation, color grade, composite, and metadata cleanup. Use when building automated video pipelines, connecting Seedance to external tools, or finishing and delivering a generated video clip.