Skip to content
ComfyDeploy
Get started

Voice Cloning with Chatterbox

Load a source recording and a reference voice, and Chatterbox converts the speech to sound like the reference. A text-to-speech branch is included for typed text.

Run on ComfyDeployDownload JSON
Voice Cloning with Chatterbox example output

How it works

This is a small audio-only workflow. It uses the audio-separation-nodes-comfyui package plus the Chatterbox nodes, and loads no image or video models.

Three LoadAudio nodes bring in the source recording and reference audio. AudioCrop trims a clip to the part you want, and AudioSeparation splits vocals from the rest so the voice is isolated before conversion.

The core step is Chatterbox_VC, which converts the source speech to the reference voice on the cuda device. The graph also contains a Chatterbox_TTS node with an exposed text field, with the sample text from the ERNIE 4.5 documentation, with a few numeric controls. In the saved workflow that node is bypassed, so you would enable it to generate speech from typed text. SaveAudio writes the results and PreviewAudio lets you listen.

Tips

  • Use a clean reference clip of a single speaker with little background noise.
  • Run the audio separation step first when your source has music or noise.
  • Crop to a short, representative section when testing.
  • To use text-to-speech, enable the bypassed TTS node and give it a reference clip.

What you can use it for

  • Re-voice a recording
  • Narration in a reference voice
  • Clean vocals before cloning
  • Prototype voices for video projects

Run it in three steps

  1. Open the template

    Click Run on ComfyDeploy. The workflow, its custom nodes and models are set up for you on a cloud machine.

  2. Add your inputs

    Provide source audio and reference voice audio in the playground or the ComfyUI editor.

  3. Run, share or deploy

    Run it on an L40S, share it with your team as a playground, or deploy it as an API endpoint.

Frequently asked questions

What GPU does this voice cloning workflow need?

It is set to run on an L40S GPU on ComfyDeploy. Both Chatterbox nodes are set to the cuda device.

Does it do text-to-speech?

The graph includes a Chatterbox_TTS node with a text input, but it is bypassed in the saved version. The active path is voice conversion with Chatterbox_VC.

How do I run it as an API?

Deploy it on ComfyDeploy and send the audio inputs to the endpoint. It is a small 13-node graph using one custom package.

Related workflows

All workflows