ComfyUI Node
LongCat AudioDiT Multi-Speaker TTS
LongCat-AudioDiT Multi-Speaker TTS. Synthesizes a conversation between multiple cloned voices. Connect reference audio clips and use [speaker_N]: tags in text.
LongCat AudioDiT Multi-Speaker TTS
- audio
◄model_path▾►
◄text[speaker_1]: Hello, I'm speaker one.
[speaker_2]: And I'm speaker two!►
◄steps16►
◄guidance_strength4.0►
◄guidance_methodapg►
◄deviceauto►
◄dtypeauto►
◄attentionauto►
◄seed0►
◄keep_model_loadedtrue►
◄pause_after_speaker0.4►
◄num_speakers▾►
CategoryLongCat-AudioDiT
Inputs (12)
| Name | Type | Default | Description |
|---|---|---|---|
| model_path | COMBO | LongCat-AudioDiT model. Models are stored in ComfyUI/models/audiodit/ | |
| text | STRING | [speaker_1]: Hello, I'm speaker one. [speaker_2]: And I'm speaker two! | Multi-speaker text. Use [speaker_1]:, [speaker_2]:, ... to assign lines to each speaker. |
| steps | INT | 164–64 | Number of ODE Euler steps. More = better quality but slower. |
| 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. 'auto' picks CUDA > MPS > CPU. |
| dtype | COMBO | auto | Model dtype. 'auto' picks bf16 for CUDA. |
| 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. |
| pause_after_speaker | FLOAT | 0.40–2 | Seconds of silence to add after each speaker turn. |
| num_speakers | COMBO | How many speakers (2-10). Changing this shows/hides speaker audio inputs. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| audio | AUDIO | — |