ComfyUI Node
Qwen3-TTS Voice Design ⚡
Generate a voice from a text description (instruct) and speak the text with it. Uses the *-VoiceDesign 1.7B checkpoint, loaded in-node.
Qwen3-TTS Voice Design ⚡
- audio
◄textHello world►
◄instruct►
◄model_choice1.7B►
◄deviceauto►
◄precisionbf16►
◄languageAuto►
◄seed0►
◄max_new_tokens2048►
◄top_p1.00►
◄top_k50►
◄temperature0.9►
◄repetition_penalty1.05►
◄attentionauto►
◄unload_model_after_generatefalse►
Category🤖 CCTech/Qwen TTS
Inputs (14)
| Name | Type | Default | Description |
|---|---|---|---|
| text | STRING | Hello world | — |
| instruct | STRING | — | |
| model_choice | COMBO | 1.7B | 2 options: 0.6B, 1.7B |
| device | COMBO | auto | 5 options: auto, cuda, xpu, mps, cpu |
| precision | COMBO | bf16 | 2 options: bf16, fp32 |
| language | COMBO | Auto | 11 options: Auto, Chinese, English, Japanese, Korean, French, +5 |
| seedopt | INT | 00–18446744073709550000 | — |
| max_new_tokensopt | INT | 2048512–4096 | — |
| top_popt | FLOAT | 1.000–1 | Nucleus sampling probability |
| top_kopt | INT | 500–100 | Top-k sampling parameter |
| temperatureopt | FLOAT | 0.90.1–2 | Sampling temperature |
| repetition_penaltyopt | FLOAT | 1.051–2 | Penalty for repetition |
| attentionopt | COMBO | auto | Attention implementation |
| unload_model_after_generateopt | BOOLEAN | false | Unload model from memory after generation |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| audio | AUDIO | — |