How to use the Seedance Camera Direction agent skill

What the seedance-camera skill makes the agent do

When you ask for a camera move, the agent picks one clear move rather than stacking several. It states the start frame, the movement, the speed, the subject relationship and the endpoint, then writes the result as a ready-to-use prompt sentence.

The skill also avoids moves that fight each other, such as a drone rise, dolly-in, handheld shake and orbit in the same short shot. For image-to-video, it aims to keep the source image composition unless you ask for a reframe. For reference videos, it can state whether a clip transfers camera movement, action rhythm or blocking, and by default it does not transfer identity.

• Chooses one primary camera move and moves extra ideas into optional variants
• Writes a physically possible phrase tied to the subject's action
• Keeps camera direction consistent across continuation clips
• Suggests splitting a shot into separate clips when it needs several beats

Install the seedance-camera skill

Download the seedance-camera ZIP from this page and unzip it. The skill is a single file, SKILL.md. To make it available to Claude Code for your user account, place the folder in ~/.claude/skills/. To limit it to one project, place it in that project's .claude/skills/ folder.

If you use the Claude apps instead, open Settings, go to Capabilities, and upload the skill under Skills.

• Personal install: ~/.claude/skills/ (available across your projects)
• Project install: .claude/skills/ inside the 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 load when you ask about camera movement, shot scale, lens feel, framing, one-take direction, dolly, pan, tilt, push-in, handheld, aerial, macro or camera transfer for Seedance 2.0. Mention Seedance 2.0 in your request so the agent knows which model you are writing for.

Describe what the viewer should feel or see, and the agent will translate that into camera language. You can also ask for a rewrite of a weak camera phrase.

• "Write a slow dolly-in for Seedance 2.0 as Character A lowers the envelope."
• "Give me a handheld shot for a Seedance 2.0 prompt, subtle and centered on the subject."
• "Rewrite this camera line for Seedance 2.0 so it has one clear move and an endpoint."

Examples of camera phrases the skill produces

The skill's reference table pairs a strong phrase with a vague one to avoid. Strong phrases name the start, the move, the speed and the endpoint. Weak phrases such as "dramatic cinematic zoom" or "cool close-up details" give the model nothing concrete to follow.

• Emotional moment: slow dolly-in from medium close-up to tight close-up as Character A lowers the envelope
• Product reveal: controlled slider move from silhouette to front three-quarter hero angle, ending on the label
• Scale: low-angle crane up from boots to skyline, ending behind the character's shoulder
• Instability: subtle handheld shoulder camera, small breathing sway, subject kept centered
• Precision: locked macro shot, focus stays on the watch gears while the second hand clicks once

Choose the right camera move

The skill's move guidance is short. Use a locked-off shot for lip-sync, product identity and delicate effects. Use a dolly-in for discovery or realization, and tracking for travel, pursuit or product motion. Use orbit only when the subject stays clear from all sides, crane or drone for scale, arrival or reveal, and handheld only when realism matters more than precision.

• Locked-off: lip-sync, product identity, delicate VFX
• Dolly-in: discovery or realization
• Tracking: travel, pursuit, product motion
• Orbit: only if the subject reads clearly from every side
• Crane or drone: scale, arrival, reveal
• Handheld: when realism matters more than precision

Safety notes: scripts and network

The skill, as published, requires nothing to be installed beyond the file itself. It does not run scripts and does not use the network, according to its listing. It only shapes the wording of generation prompts. It does not edit footage or operate a real camera.

The skill refers to several other reference files, such as quick-ref, cinematography-shot-language, directing-engine and Chinese or Russian vocabulary files. Those are linked inside SKILL.md but are not part of the single-file download listed here, so the agent may not be able to load them from this package alone.

• Runs scripts: no
• Uses network: no
• Requires: nothing listed
• Licence: MIT

Tips for better camera prompts

Give the agent one shot at a time. If your brief has several beats, ask for separate clips or a time-segmented prompt. Name the subject and where it sits in frame, since the skill keeps moves tied to the subject's action.

Treat the output as a strong starting point, not a guarantee. The video model may not follow the wording exactly, so generate several takes and adjust the endpoint or speed if the result drifts.

• Name the subject and their action in the same sentence as the move
• Specify speed and endpoint, for example 'slow' and 'ending on the label'
• Lock the framing for lip-sync or product shots
• For complex moves, consider a reference video instead of a long verbal description