How to Use LatentSync in Claude Code
LatentSync matches a recorded speaker's mouth movements to replacement speech for dubbing and dialogue corrections. It edits existing footage while aiming to preserve the speaker's identity and head pose.
Prompts to try with LatentSync in Claude Code
Copy an example into Claude Code, or ask in your own words.
Use LatentSync to match assets/presenter.mp4 to assets/corrected-line.wav. Save a review cut and compare the mouth timing with the replacement recording.
Use LatentSync on the first 10 seconds of assets/lesson.mp4 with the matching passage from assets/revised-narration.wav. Prepare a comparison before processing the rest.
Use LatentSync to make two comparison cuts from assets/host.mp4 and assets/new-dialogue.wav, with guidance at 1 and 1.5. Keep the seed fixed and compare mouth timing and jitter.
Use LatentSync to match assets/character.mp4 to assets/revised-voice.wav. Add the result to the project's dialogue review page beside the original.
Use LatentSync for each presenter clip and replacement recording paired in the project's media manifest. Produce a review checklist covering lip timing, teeth, and facial stability.
How to set up LatentSync in Claude Code
Pick the way that suits you. Each takes about two minutes.
Use Spiral's MCP server
- Add the hosted server
This adds Spiral to the current project. Add --scope user to use it in every project.
claude mcp add --transport http spiral https://api.spiral.ai/mcp
- Sign in
Start claude, type /mcp, choose spiral and Authenticate. Sign in to Spiral in the browser and click Authorize. No API key.
Use the Spiral skill
- Send Claude Code this message
Your agent reads Spiral's skill and sets itself up.
set up https://spiral.ai/SKILL.md
- Sign in or paste a key
Approve Spiral in the browser, or give the agent a key from the console.
Use the Spiral CLI
- Install and sign in
Installs the Spiral CLI, signs you in and adds its skill to your coding agents.
npx -y spiral-cli@latest init
Call the Spiral API
- Create an API key
In the console, under API keys.
- Run LatentSync
One request; the result comes back by polling or a webhook.
curl https://api.spiral.ai/v1/runs \ -H "Authorization: Bearer $SPIRAL_API_KEY" \ -H "Content-Type: application/json" \ -H "Idempotency-Key: $(uuidgen)" \ -d '{ "provider": "fal", "model": "fal-ai/latentsync", "input": { "video_url": "https://fal.media/files/koala/8teUPbRRMtAUTORDvqy0l.mp4", "audio_url": "https://fal.media/files/lion/vyFWygmZsIZlUO4s0nr2n.wav" } }'
Then paste one of the prompts above: Claude Code finds LatentSync on Spiral, shows the price and runs it.
LatentSync providers and prices
Spiral runs LatentSync through fal. Open a card for today's price, with every fee included.
Settings you can ask for
LatentSync works from your video URL and audio URL. Ask Claude Code for any of these in your prompt.
| Setting | Allowed values | Default | Description |
|---|---|---|---|
| Guidance scale | 1 to 2 | 1 | Guidance scale for the model inference |
| Seed | a whole number | Random seed for generation. If None, a random seed will be used. | |
| Loop mode | pingpong or loop | Video loop mode when audio is longer than video. Options: pingpong, loop |
Every API field of LatentSync is on its tool page: open its card above.
What can you do with LatentSync in Claude Code?
Turn existing project videos and replacement recordings into dialogue edits with LatentSync. A coding agent can organize comparisons and flag clips that need review for mouth timing, distortion, or jitter.
Dub a presenter clip
Match an existing presenter shot to a recorded translation. Review the mouth timing against the translated speech before editing the result into the finished video.
Correct a spoken line
Replace a mispronounced name or outdated sentence in a talking-head segment using a newly recorded line.
Update training videos
Use existing instructor footage with revised narration for a changed lesson or procedure. Start with a short passage for review.
Revise animated dialogue
Match an animated character's mouth movements to a replacement voice performance in an existing dialogue shot.
Prepare spokesperson ad edits
Create separate review cuts of an existing spokesperson video for recorded messages aimed at different audiences.
Build language lesson clips
Pair presenter footage with recorded example phrases for a language lesson. Have a fluent reviewer check the resulting mouth movements.
More on LatentSync and Claude Code
Overview
LatentSync in other agents
Questions about LatentSync in Claude Code
Can Claude Code use LatentSync?
Yes. Add Spiral to Claude Code as an MCP server, and Claude Code can find and run LatentSync like any built-in tool.
How much does LatentSync cost in Claude Code?
Up to $0.39 per run at the default inputs for 60 seconds of input. You pay from your Spiral balance as you go, with no subscription, and a failed run is not charged. Your Claude Code plan is separate.
Do I need a fal account to use LatentSync in Claude Code?
No. Spiral runs LatentSync on its own fal account. You sign in to Spiral once and never paste a key.
How long does it take to set up LatentSync in Claude Code?
About two minutes: add the hosted server, then sign in.
Does Claude Code wait while LatentSync runs?
LatentSync runs in the background. Claude Code starts the run, checks on it and gives you the video when it's ready.
What is LatentSync best used for?
Use LatentSync to match a recorded speaker's mouth to replacement dialogue. Dubbing, corrected narration, and updated presenter clips are practical starting points.
Can LatentSync work with animated characters?
Yes. LatentSync supports both real people and animated characters in existing footage; try a short dialogue shot before processing a longer scene.
Why can stronger lip synchronization make the video look worse?
Higher guidance can improve mouth-to-audio alignment while introducing distortion or jitter. Increase it gradually and compare both timing and facial stability.
Does LatentSync support Chinese speech?
ByteDance specifically improved Chinese-video performance in LatentSync 1.5 by adding Chinese training data. For Chinese dialogue, test a representative passage rather than assuming identical results across releases.
What does the published 512 by 512 resolution mean?
ByteDance describes 512 by 512 as the training resolution for LatentSync 1.6, introduced to address blurry teeth and lips. Treat that figure as release-specific training information when planning output quality.
Start using LatentSync in Claude Code
One Spiral account for every tool your agent needs. No subscription; you pay per run.
Paste this into Claude Code:
set up https://spiral.ai/SKILL.md


