ComfyUI Node
arkennemasis Qwen3-TTS (voice clone)
Qwen3-TTS, running locally on the GPU with no account. Wire a clip into `reference_audio` and it speaks your text in that voice.
arkennemasis Qwen3-TTS (voice clone)
- reference_audio
- audio
- report
◄text►
◄model▾►
◄language▾►
◄seed0►
◄reference_text►
◄timeout_seconds180►
Categoryarkennemasis/Audio
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| text | STRING | What to say. | |
| model | COMBO | A folder under ComfyUI/models/qwen-tts. Voice cloning needs a *Base* model. | |
| language | COMBO | 'Auto' lets the model decide from the text. | |
| seed | INT | 00–18446744073709550000 | — |
| reference_audioopt | AUDIO | The voice to clone. 5-30 seconds of clean speech is plenty. A *Base* model REQUIRES this. | |
| reference_textopt | STRING | A transcript of the reference clip. Supplying it gives a closer clone; leaving it blank uses x-vector-only mode, which still works. | |
| timeout_secondsopt | INT | 18060–3600 | Per attempt, and there are 3 attempts with fresh seeds. A 24-word line takes about 30 s including the ~2.5 GB weight load, so 180 s is six times over — it is a runaway detector, not a budget. It was 600 s, which meant one looping line cost 10 minutes before failing. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| audio | AUDIO | — |
| report | STRING | — |