How to use the Seedance Sequence Planner agent skill

What the Seedance Sequence Planner skill makes the agent do

When a request is larger than one reliable generation, the skill makes the agent plan the whole story first. It sets the story promise and final outcome, orders the beats, groups them into scenes, and divides each scene into generation-sized clips.

Only the next unresolved clip gets a finished prompt. Later clips stay as provisional intent cards until the earlier footage exists. The agent tracks what has already happened, what must not happen yet, and how each clip should end so that characters and looks stay consistent across generations. The skill's own summary is: plan globally, generate locally.

• Plans the whole story and its scenes before writing any clip prompt
• Writes a prompt only for the next clip, based on the accepted footage
• Records each clip's start and end state so the next clip can continue from it
• Decides when to cut to a new scene instead of chaining more footage

How to install the skill

Download the skill ZIP from the skill's page in this directory. Then choose one of the two install routes below.

Option A is a personal install: unzip the file into ~/.claude/skills/. Option B is a project install: unzip the file into the .claude/skills/ folder of your project. The skill's source repository is github.com/Emily2040/seedance-2.0.

Option C is to upload the skill in the Claude apps. Open Settings, go to Capabilities, then Skills, and upload the ZIP there.

• Personal: ~/.claude/skills/
• Project: .claude/skills/ inside your project folder
• Claude apps: Settings → Capabilities → Skills

How to trigger the skill with natural requests

You do not need a special command. The skill is meant to be triggered by describing the kind of job you have. Its description says it should be used when a Seedance 2.0 request is a long story, a connected set of clips, a multi-generation scene, a campaign sequence, a dense storyboard, a continuation-ready plan, or any idea that must be divided into stateful clips.

Phrases that also trigger it include continue, extend, next part, part two, next scene, and make it longer. These are useful when you already have one clip and want the next one to match it.

• Plan a 3-part ad campaign for Seedance 2.0 with the same presenter in every clip
• Write the next part of this story, keeping the same characters and location
• Make it longer: a short film about a delivery rider, split into connected clips
• Part two should start where part one ends

An example of how a project unfolds

Suppose you describe a five-minute story. The skill does not treat it as one long extension chain. According to the skill, a story like this usually resolves to several scenes of two to five clips each. Each scene has one location and time envelope and opens from canonical reference material.

Within a scene, clips can chain from the accepted footage of the previous clip, but only up to a limit. The default is two consecutive output-sourced generations, with a hard ceiling of three. After that, the plan schedules a re-anchor, because identity drifts over chained generations. Cutting to a new scene is the cheapest way to keep continuity, since audiences expect continuity only inside a shot.

• Generate a clip, then check the footage that was actually produced
• Record the observed end state before writing the next prompt
• Re-anchor at a scheduled scene boundary rather than repairing drift later

What you need before you start and the safety notes

You need Seedance 2.0 access to generate the clips. The skill writes prompts; it does not generate video itself. The skill is also tailored to Seedance 2.0's prompt syntax, durations, and reference tags, so it is not a general video tool.

The skill loads reference files from the same repository, including directors-read, prompt-compiler, and continuity-qc. These reference files were not included in the files provided for this listing, so the skill may not work fully without them. The skill runs no scripts and makes no network requests.

Audio is not continuous between generations. Clips carry ambience, sync sound effects, and on-camera dialogue, but music and score should be unified in post-production.

• Requires Seedance 2.0 access to generate clips
• Reference files from the repository are needed for full behaviour and were not included here
• Runs scripts: no
• Uses network: no
• Music must be unified in post-production

Tips for better results

Give the agent the full arc up front, including the ending. The plan is built from the final outcome, so a clear ending makes every clip easier to place.

When a clip is accepted, keep the footage and share it. The next prompt is built from what the footage actually shows, not from what was intended.

• State the final outcome and the number of clips you want
• Name the recurring characters, props, and locations
• Keep single short clips on the standard prompt path, since the planning overhead is not needed
• Plan music separately and add it in post-production