How to use the Seedance 2.0 Video Prompt Writer agent skill

What the seedance-prompt skill makes the agent do

An agent skill is a folder containing a SKILL.md file that tells an AI agent how to handle a specific kind of task. Once this skill is installed, the agent turns a rough video idea or a set of reference assets into a tight prompt for the Seedance 2.0 video model.

Each prompt specifies the subject, the visible change, one clear camera move, lighting, audio, and what must stay consistent. The agent also picks the right mode for the job, and it can compress, translate, and debug prompts you already have.

• Writes new prompts from a concept, a rough idea, or reference assets
• Improves, compresses, translates, or debugs existing prompts
• Chooses a mode: text-to-video, image-to-video, video-to-video, multi-reference, or first/last frame
• Loads supporting reference guides for multi-shot plans and non-English prompts

Install the seedance-prompt skill

Download the ZIP file from this page and unzip it. The skill is a single file, SKILL.md, inside the seedance-prompt folder. Place that folder in one of two locations, depending on how you want to use it.

For a personal setup available across your projects, put it in ~/.claude/skills/. For a setup that applies to one project only, put it in the project's .claude/skills/ folder.

If you use the Claude apps instead of a folder on your computer, open Settings, go to Capabilities, then Skills, and upload the skill there.

• Personal: ~/.claude/skills/seedance-prompt/
• Project only: .claude/skills/seedance-prompt/ inside your project folder
• Claude apps: Settings → Capabilities → Skills, then upload

Trigger the skill with natural requests

You do not need a special command. The skill is meant to be used when you ask the agent to write, improve, translate, compress, or debug a Seedance 2.0 video prompt. It also responds to mentions of T2V, I2V, V2V, R2V, camera direction, prompt quality, or reference assets you want to use for a production-ready prompt.

Describe what you want to see on screen and say what you have. The more concrete the description, the fewer rounds of revision you need.

• Write a Seedance 2.0 prompt for a product shot: a ceramic perfume bottle with condensation forming on the glass.
• Turn these three product photos into an image-to-video prompt that keeps the label exactly as it is.
• This Seedance prompt drifts after two seconds. Debug it and tell me what to change.
• Compress this prompt to fit the length limit without losing the camera move.
• Translate this Seedance prompt into Japanese and keep the shot order.

Pick the right mode and read a prompt

The skill chooses a mode before drafting. Text-to-video (T2V) builds the whole shot from words. Image-to-video (I2V) starts from your image and adds only motion, timing, camera, lighting change, audio, and preservation rules. Video-to-video (V2V) uses a source clip for camera move, rhythm, or blocking. Reference-to-video (R2V) assigns a separate role to each asset. First/last frame (FLF2V) moves from a first image to a last image. Edit changes one layer of an existing clip, and Extend continues from footage you have already accepted.

A prompt follows the pattern Subject, Action, Scene, Camera, Lighting and style, Audio, Constraints. The subject and action come first because early clauses set what the model focuses on. The example below uses phrases from the skill's own examples.

What you need before you start

The skill only writes prompts. It does not generate video. To produce the clips, you need your own Seedance 2.0 video generation account. You do not need that account to write or revise prompts.

The agent must be able to read SKILL.md skill files. The skill runs no scripts and does not use the network, so the only thing it does is read and write text within your agent.

• An AI agent that can read SKILL.md skill files
• A Seedance 2.0 video generation account, only if you want to produce clips
• No scripts are run by this skill
• No network use by this skill

Safety notes on references and likeness

The skill itself warns you to use owned or authorised reference material. Do not ask it to recreate a real person's likeness or a copyrighted scene from reference clips unless you have the rights to do so.

When you use video-to-video or reference images, tell the agent which parts should transfer and which should not. The skill's guidance is to restrict what each reference is allowed to control, for example limiting a clip to camera movement or timing rather than identity.

• Use images and clips you own or are licensed to use
• Say which reference controls identity, motion, camera, or style, and which must not transfer
• Review generated clips before publishing them

Tips for better prompts

Give the agent one visible change per clip. The skill's guidance is to keep one visible beat and one endpoint in a single shot. If you have several events, ask for a storyboard, where the skill writes numbered shots in event order.

When you want a revision, ask for one change at a time. Each revision keeps the decisions already made and changes only what you asked for. Use phrases such as preserve the reference exactly, or change only the lighting.

• Keep one camera move per shot, with a clear start and end
• For product clips, name the labels and logos that must be preserved
• If a prompt is too long, ask the agent to compress it and keep the camera move
• Ask for debugging when a clip drifts, overloads the shot, or breaks its length limit