How to use the Seedance Audio Director agent skill

What is an agent skill, and what does this one do?

An agent skill is a folder containing a SKILL.md file. The file gives an AI agent specific instructions for one kind of task, and the agent loads it when a request matches the skill's description. This skill is named seedance-audio and is meant for the audio side of Seedance 2.0 video prompts.

When active, the agent names the speaker, quotes the exact dialogue, and lays out sound effects, music, ambience and silence. It also checks timing so that the audio supports what is visible on screen rather than competing with it. The skill additionally covers fixing lip-sync drift, mapping audio reference files to roles, and handling multiple speakers.

• Speaker map and quoted dialogue, kept in the user's original language and wording
• Sound layers: Dialogue, Sound, SFX, Music and Silence, including only the layers that matter
• Audio reference roles, such as using an audio file for tempo only
• Lip-sync constraints and desync repair suggestions
• Post-production and delivery notes when they are needed

What you need before you start

You need two things. The first is a Seedance 2.0 video generation account or tool to run the prompt, because the skill only writes direction and does not generate video itself. The second is Claude Code, Claude, or another agent environment that reads SKILL.md-compatible skills.

The skill does not run scripts and does not use the network, according to its own metadata. It only supplies written guidance to the agent.

How to install the seedance-audio skill

Download the skill ZIP from the download link on this page. Then choose one of the two installation routes below.

For Claude Code, unzip the file into your personal skills folder at ~/.claude/skills/ so it is available in every project. To limit it to one project, unzip it into that project's .claude/skills/ folder instead.

In the Claude apps, upload the ZIP under Settings → Capabilities → Skills.

• Personal install: ~/.claude/skills/ (available across projects)
• Project install: .claude/skills/ inside your project 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 is meant to be used when you ask for Seedance 2.0 audio, dialogue, lip-sync, music, sound effects, ambience, beat-sync, audio-reference mapping, desync troubleshooting, or sound-driven visual timing. Describe the clip in plain language and name the sound you want.

Give the agent your scene, the exact line if there is dialogue, and any reference files you plan to upload. The more specific the request, the more usable the output.

• Write the scene and the exact dialogue you want spoken, with the speaker's name
• Say which sound matters most: lip-sync, music, a sound effect or ambience
• Name any audio reference and what it should control, such as tempo only

Examples of prompt layers the skill produces

The skill uses compact layers in the form Dialogue, Sound, SFX, Music and Silence. Each layer is optional. Silence is valid when it sharpens a moment or avoids confusing lip-sync. The examples below show the style the skill writes in.

• Lip-sync: a locked medium close-up where Character A says the quoted line with clear dry dialogue and no head turn
• Product ad: low room tone, a magnetic click on lid open, and a soft glass chime at the final frame
• Beat sync: an audio file provides tempo only, and light pulses and foot taps match the downbeat
• Drama: distant rain and refrigerator hum with no music during the line
• Action: breathing that grows louder, a shoe squeak at landing, and a metal door buzzer at the endpoint

Tips for better lip-sync and fewer timing problems

The skill recommends stable framing and one speaker turn per short clip when mouth accuracy matters. For multiple characters, it advises separate turns and a stable camera, or generating single-speaker clips and combining them in post. Check the spoken duration of your line and review the returned words, performance and visible sync as separate checks.

If dialogue drifts, the skill suggests a shorter line if the wording can change, a locked camera, no head turns, a cleaner audio reference and fewer competing sound effects. It also warns that outcomes cannot be guaranteed, so plan for a few generation attempts.

• Use one speaker per short clip when reliability matters
• Lock the camera and remove head turns
• Reduce competing SFX and extra music instructions
• Make the audio reference role explicit, for example: @Audio1 controls tempo and energy
• Mute a reference video before upload when possible if audio and video references conflict

Safety notes on voices, rights and scope

The skill does not edit or mix real recorded audio. It only writes generation direction and post-production notes. It also states that cloning a real person's voice requires rights clearance, so only use voice material you have the rights to use.

Spoken-voice reference workflows are used only when the active operation supports them. The skill does not promise exact playback or guaranteed lip-sync.

• Not for editing or mixing real recorded audio
• Voice references must be rights-cleared
• No guaranteed lip-sync results