ComfyUI Node
LongCat AudioDiT Voice Clone TTS
LongCat-AudioDiT Voice Clone TTS. Clones voice from reference audio using diffusion-based generation.
LongCat AudioDiT Voice Clone TTS
- prompt_audio
- audio
◄model_path▾►
◄textThe sun glows warmly in a cloudless blue sky, a soft breeze drifts through the air, and birds fill the world with gentle, cheerful songs. Everything feels alive with beauty, just waiting to be discovered.►
◄prompt_text►
◄steps16►
◄guidance_strength4.0►
◄guidance_methodapg►
◄deviceauto►
◄dtypeauto►
◄attentionauto►
◄seed0►
◄keep_model_loadedtrue►
CategoryLongCat-AudioDiT
Inputs (12)
| Name | Type | Default | Description |
|---|---|---|---|
| model_path | COMBO | LongCat-AudioDiT model. Models are stored in ComfyUI/models/audiodit/ | |
| text | STRING | The sun glows warmly in a cloudless blue sky, a soft breeze drifts through the air, and birds fill the world with gentle, cheerful songs. Everything feels alive with beauty, just waiting to be discovered. | Text to synthesize in the cloned voice. |
| prompt_audio | AUDIO | Reference audio to clone the voice from. 3-15 seconds gives the best results. | |
| prompt_text | STRING | Transcript of the prompt audio. Required for voice cloning. Improves quality significantly. | |
| steps | INT | 164–64 | Number of ODE Euler steps. |
| guidance_strength | FLOAT | 4.00–10 | CFG/APG guidance strength. |
| guidance_method | COMBO | apg | Guidance method. 'apg' recommended for voice cloning. |
| device | COMBO | auto | Compute device. |
| dtype | COMBO | auto | Model dtype. |
| attention | COMBO | auto | Attention implementation. |
| seed | INT | 00–2147483647 | Random seed. 0 = random. |
| keep_model_loaded | BOOLEAN | true | Keep model loaded between runs. Model is automatically offloaded to CPU after generation to free VRAM, then resumed to GPU on the next run. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| audio | AUDIO | — |