ComfyUI Node
WavTTS Generate
Generate ComfyUI AUDIO from text using WavTTS zero-shot voice prompting.
WavTTS Generate
- wavtts_model
- reference_audio
- audio
◄textHello! This is WavTTS running inside ComfyUI.►
◄reference_text►
◄Steps50►
◄CFG3.0►
◄speed1.00►
◄timestep_mappingpower►
◄timestep_power2.0►
◄shift3.0►
◄cross_fade_seconds0.00►
◄fixed_total_duration_seconds0.0►
◄seed0►
CategoryWavTTS
Inputs (13)
| Name | Type | Default | Description |
|---|---|---|---|
| wavtts_model | WAVTTS_MODEL | Loaded WavTTS model from the WavTTS Load Model node. | |
| reference_audio | AUDIO | Reference speaker audio. Use a clean clip under 12 seconds. | |
| text | STRING | Hello! This is WavTTS running inside ComfyUI. | Text to synthesize. |
| reference_text | STRING | Transcript of the reference audio. Required for WavTTS voice prompting. | |
| Steps | INT | 504–128 | Number of WavTTS sampling steps. Higher can improve quality but is slower. |
| CFG | FLOAT | 3.00–10 | Classifier-free guidance strength. Higher values follow the text harder, but can sound less natural. |
| speed | FLOAT | 1.000.1–3 | Speech pace multiplier used for duration estimation. 1.0 is normal. |
| timestep_mapping | COMBO | power | Sampler timestep schedule. power is the upstream default; sway_sampling enables the EPS/Sway path. |
| timestep_power | FLOAT | 2.00.1–8 | Exponent for power timestep mapping. Only meaningful when timestep_mapping is power. |
| shift | FLOAT | 3.00.1–8 | Flow timestep shift. Lowering this can help if speech contains background noise. |
| cross_fade_seconds | FLOAT | 0.000–2 | Overlap crossfade used only when long text is split into multiple generated chunks. |
| fixed_total_duration_seconds | FLOAT | 0.00–120 | 0 lets WavTTS estimate duration. Positive values force total prompt plus generated duration. |
| seed | INT | 00–2147483647 | 0 is random/unseeded. Positive values make generation repeatable. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| audio | AUDIO | — |