ComfyUI Node
StepAudioEditX - Clone 🎤
A ComfyUI node in audio/step_audio with 14 inputs and 1 output.
StepAudioEditX - Clone 🎤
- prompt_audio
- audio
◄prompt_text►
◄target_text►
◄model_path(No models found - download Step-Audio-EditX)►
◄devicecuda►
◄torch_dtypebfloat16►
◄quantizationnone►
◄attention_mechanismsdpa►
◄temperature0.7►
◄do_sampletrue►
◄max_new_tokens4096►
◄longform_chunkingfalse►
◄seed0►
◄keep_model_in_vramtrue►
Categoryaudio/step_audio
Inputs (14)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt_text | STRING | Exact transcript of what the reference audio says. Must match the audio content perfectly for best voice cloning results. The AI uses this to align the voice characteristics. | |
| target_text | STRING | The new text you want to speak in the cloned voice. Can be any text - the AI will speak it with the reference voice's tone, accent, prosody, and speaking style. | |
| model_path | COMBO | (No models found - download Step-Audio-EditX) | Step-Audio-EditX model location. Place models in ComfyUI/models/step_audio/ or select from discovered models. |
| device | COMBO | cuda | Hardware device: 'cuda' for GPU (10-100x faster, ~8GB VRAM) or 'cpu' (very slow, no VRAM needed). Always use CUDA if available. |
| torch_dtype | COMBO | bfloat16 | Model precision: bfloat16 (best quality, stable, 8GB VRAM), float16 (good quality, 6GB VRAM), float32 (max quality, 16GB VRAM), auto (selects best for your GPU). |
| quantization | COMBO | none | VRAM reduction: 'none' (best quality, 8GB VRAM), int8 (good quality, 4GB VRAM), int4 (acceptable quality, 3GB VRAM), int4_awq (optimized int4, requires pre-quantized model). Use if low on VRAM. |
| attention_mechanism | COMBO | sdpa | Attention layer: sdpa (fastest, good VRAM, default), eager (slowest, most stable), flash_attn (fastest, needs RTX 30xx+), sage_attn (best VRAM efficiency). |
| temperature | FLOAT | 0.70.1–2 | Voice variation control: 0.1-0.5 (very consistent, robotic), 0.6-0.8 (natural, recommended), 0.9-1.5 (varied, expressive), 1.6-2.0 (very random, creative). Controls randomness in speech generation. |
| do_sample | BOOLEAN | true | Sampling mode: True (natural, varied speech with temperature control) or False (deterministic, uses greedy decoding, ignores temperature). Keep True for natural-sounding voices. |
| max_new_tokens | INT | 4096256–16384 | Maximum audio tokens to generate: 2048 (~10s), 4096 (~20s), 8192 (~40s). Higher = more VRAM + time. With longform chunking enabled, each chunk is limited to this value. |
| longform_chunking | BOOLEAN | false | Smart text splitting for long content (>2000 words). Splits at sentence boundaries, generates chunks separately, then stitches seamlessly. Slower but handles unlimited length. Disable for short text. |
| seed | INT | 00–18446744073709550000 | Reproducibility seed: 0 (random output each time) or any number (same input = same output). Use fixed seed to reproduce exact voice outputs with identical parameters. |
| keep_model_in_vram | BOOLEAN | true | Model caching: True (keep loaded in VRAM, fast repeated use) or False (unload after use, frees ~8GB VRAM). Disable if switching between multiple models frequently. |
| prompt_audioopt | AUDIO | Reference voice audio (3-30 seconds recommended, 0.5-30s supported). The AI will analyze and clone this voice's tone, accent, pitch, rhythm, prosody, and speaking style for target_text generation. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| audio | AUDIO | — |