Guide · Agent skill
How to use the Seedance Character Consistency agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The seedance-characters skill is an instruction file for an AI agent that writes video-generation prompts featuring people or characters. It tells the agent to give each character a stable tag, track their wardrobe, position and expression, and keep scenes with several people simple so faces and bodies do not drift between shots. After this guide you will know how to install the skill, how to ask for it in plain language, and what it will and will not do.
Seedance Character ConsistencyEmily2040/seedance-2.0 · MIT Download the skill Quick start
- 1Download the ZIP. Download the seedance-characters skill ZIP from this page.
- 2Unzip into a skills folder. Extract it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder for one project.
- 3Or upload in the Claude apps. In the Claude apps, open Settings, go to Capabilities, then Skills, and upload the ZIP.
- 4Describe your characters. Name each character, the number of people in the shot and what must stay the same.
- 5Ask in plain language. Request character consistency, multi-character blocking or wardrobe continuity in ordinary words. The agent will apply the skill.
- 6Check the output. Confirm the character card, tag map, action assignments and continuity constraints before you paste the prompt into your video tool.
What is an agent skill and what does this one do?
An agent skill is a folder containing a SKILL.md file. When the skill is installed and relevant, the AI agent reads these instructions and follows them while it works. This skill covers character consistency for Seedance 2.0 prompts.
The skill makes the agent assign each character a stable tag such as Character A or @Image1 subject, then keep that character's role, appearance, wardrobe, position, action and emotional beat consistent. When more than one person is on screen, it asks the agent to assign actions separately so the video model knows who moves and who stays still.
- Assigns stable character tags and avoids ambiguous pronouns in multi-character scenes
- Tracks wardrobe, hair, screen position, eyeline and expression across a multi-clip sequence
- Limits large movements in crowded shots and keeps hands and faces simple
- Advises caution with real people's likenesses and suggests original character archetypes when permission is unclear
Install the seedance-characters skill
Download the skill's ZIP file from this page. Unzip it into your personal skills folder at ~/.claude/skills/, or into a project's .claude/skills/ folder if you want the skill to apply only to that project.
If you use the Claude apps instead, open Settings, go to Capabilities, then Skills, and upload the ZIP there. The skill's package contains one file, SKILL.md, so there is nothing else to configure.
- Personal install: ~/.claude/skills/
- Project install: the project's .claude/skills/ folder
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
How to trigger the skill with natural requests
You do not need a special command. The skill's description tells the agent to use it when you ask about character consistency, character tags, identity lock, multi-character blocking, wardrobe continuity, hand safety, expression control or likeness-sensitive character guidance. Describe what you need in ordinary words.
The more specific your request, the more useful the output. Name the characters, how many clips there are and what must stay the same from clip to clip.
- 'Write a Seedance 2.0 prompt with two original characters arguing in a kitchen. Give them stable tags and keep their wardrobe the same.'
- 'I have a four-clip sequence. Keep Character A's blue jacket, hair and eyeline consistent across every shot.'
- 'Block a three-person scene so only one person reacts and the others stay in subtle micro-motion.'
- 'Help me keep hands simple in a close-up of Character B during dialogue.'
Use the character contract and blocking rules
The skill asks the agent to fill in a character contract for each person: tag, identity anchor, position, one action with an endpoint, expression as observable behavior, and a constraint listing what must stay unchanged. Write actions as concrete verbs, not vague moods.
For multi-person shots, the skill uses a three-tier action hierarchy. Everyone not in focus gets persistent micro-motion such as breathing, blinks or drifting gaze. One person may get a single focused response with a time window. Large actions such as standing, walking, turning or picking up an object are excluded by default unless they are the shot's only beat.
- Persistent micro-motion: the default for everyone who is not the focus
- One focused response: one small reaction with an explicit time window
- Large actions: excluded by default in multi-person shots
Keep multi-clip sequences consistent
When a character must look the same across several shots, the skill points the agent to a shot-list continuity reference. The agent then carries forward wardrobe, hair, screen geography, eyeline, pose, emotional state and continuity locks from one clip to the next.
Canonical identity references control who the character is. Accepted footage controls the state at the start of a new clip. Locked details should not be overwritten by a later motion instruction.
- Carry wardrobe, hair, props, eyeline and screen direction between clips
- Treat each recurring character as a continuing cast member
- Lock the camera for lip-sync or portrait-preservation shots
Safety, permissions and what the skill does not check
The skill bundle does not run scripts and does not use the network. It produces text only, so you need Seedance 2.0 or another video generation tool to turn the prompts into footage.
For real people, the skill tells the agent not to infer consent from an uploaded image or voice. If authorization is unclear, the agent should rewrite the character as an original archetype while keeping the scene's purpose. The skill only advises caution. It does not verify whether you have consent, so you remain responsible for permissions.
- Runs scripts: no
- Uses network: no
- Requires a separate video generation tool to use the prompts
- Does not verify consent for real-person likenesses
- Not suited to static illustration or written fiction, where the blocking and shot rules do not apply
What people use it for
- Writing Seedance 2.0 prompts for a short series with recurring original characters
- Blocking dialogue scenes with two or more people so only one person reacts at a time
- Keeping a character's jacket, hair and eyeline consistent across a multi-clip sequence
- Simplifying hand and face actions in close-ups where fine detail tends to degrade
Frequently asked questions
Is the seedance-characters skill free to use?
The skill is published under the MIT licence, as stated on this page. The MIT licence generally allows use, modification and redistribution, but check the repository's licence file for full terms. Seedance 2.0 or another video tool is a separate requirement.
What do I need to use the seedance-characters skill?
You need an AI agent that supports skills, such as Claude, and a video generation tool like Seedance 2.0 to consume the prompts. The skill itself only produces text and does not run scripts or use the network.
Where do I install a Claude agent skill?
Unzip the skill into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, you can instead upload the ZIP under Settings, then Capabilities, then Skills.
Why does my character's face or outfit change between clips?
Models tend to drift when scenes are crowded or instructions are vague. Give each character a stable tag, name the wardrobe and constraints explicitly, and keep the number of large actions low. The skill is designed to help with exactly these details.
Can I use this skill to recreate a real person's face or voice?
The skill advises caution with real-person likenesses and tells the agent to switch to an original archetype when permission is unclear. It does not verify consent, so you must have clear authorization yourself before using a real person's likeness.
How is this different from writing prompts without a skill?
Without a skill, you must remember every rule yourself. This skill gives the agent a consistent method: stable tags, a character contract, a three-tier action hierarchy and continuity locks. It does not replace a video generation tool.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you produce short AI video with recurring original characters, write Seedance 2.0 prompts for scenes with two or more people, or need consistent wardrobe and eyeline across clips. Pick something else if you need to generate or clone a real person's face or voice without verified authorization, or if your work is static illustration or written fiction.
Download the skill