Skip to content
BytePlusBytePlusBytePlus ModelArk for ComfyUI

Seed Voice Clone

Template
Seed Voice Clone, in the ComfyUI template browser under BytePlus
Workflow file
Seed Voice Clone.json
BytePlus nodes
Seed Voice Clone, Seed Speech TTS
Keys needed
Seed Speech API key
Also uses
Load Audio, Preview Audio (built into ComfyUI)

What it makes

Section titled “What it makes”

Two audio clips. Seed Voice Clone trains a voice from your reference clip and returns a demo clip that says “Hello! This is a preview of my cloned voice.” Seed Speech TTS then speaks “Now I can say anything in my own voice.” with the cloned voice, using model seed-icl-2.0. The voice is identified by a speaker ID that the clone node passes to the TTS node.

The workflow

Section titled “The workflow”
The Seed Voice Clone template: Load Audio into Seed Voice Clone, whose speaker ID feeds Seed Speech TTS, with a Preview Audio after each
Five nodes: Load Audio, Seed Voice Clone, Seed Speech TTS and two Preview Audio.
  1. Load Audio supplies the reference clip. The template ships with no file chosen.
  2. Seed Voice Clone trains the voice into your voice slot, with language en, and waits until it is ready. Its demo_audio goes to a Preview Audio.
  3. Seed Speech TTS receives the speaker_id as custom_speaker_id and speaks the new sentence with seed-icl-2.0.
  4. The second Preview Audio plays the result.

Run it

Section titled “Run it”
  1. Add your Seed Speech API key in Settings → BytePlus if you have not yet (API keys). This is a separate key from the ModelArk key.
  2. Open the template browser and choose Seed Voice Clone under BytePlus.
  3. In Load Audio, choose a clip of 10 to 15 seconds of clear speech, up to 10 MB.
  4. In Seed Voice Clone, replace the placeholder in speaker_id with your voice slot ID (starts with S_, bought in the Seed Speech console) or your own postpaid custom voice ID.
  5. Set language to the language spoken in the clip. Cross-language cloning is not supported, so write the TTS text in the same language.
  6. Press Run. The demo clip and the new sentence appear in the two Preview Audio nodes.

Settings worth changing

Section titled “Settings worth changing”
  • speaker_id: the voice slot to train into. Each slot can be trained 15 times.
  • reference_text: optional, the words spoken in the clip. Training fails if the audio differs too much from it.
  • demo_text: optional text for the demo clip, 4 to 80 characters in the same language.
  • text (TTS): the sentence to say in the cloned voice.
  • tone_fidelity (TTS): stays as close as possible to the training audio’s voice, emotion and accent, for the same language only.

Example result

Section titled “Example result”

Replace S_your_voice_slot with a valid voice slot from Seed Speech, then run the template to create your voice and hear the TTS result.

Time and cost

Section titled “Time and cost”

Seed Voice Clone trains into a voice slot from the Seed Speech console. Every run with changed inputs trains the slot again, and a slot can be trained 15 times. According to the pack’s README, the first TTS call with the voice starts the slot’s billing. The TTS seed control is set to randomize, so every run speaks the sentence again; set it to fixed and ComfyUI reuses the cached result of an unchanged node.

Section titled “Related”