ComfyUI-VibeVoice

Why we included this project

ComfyUI users who already build audio workflows will find this node a convenient way to bring Microsoft's VibeVoice into their graphs. It generates expressive, long-form conversational audio with up to four distinct voices in a single output, and it handles the fiddly parts, model downloads, VRAM management, and audio processing, so you can focus on the script. The hybrid mode is worth calling out: you can clone one speaker's voice from a wav or mp3 reference while the model invents fresh zero-shot voices for the rest of the same dialogue. Scripting accepts either [1] tags or the classic 'Speaker 1:' format, and the attention options plus 4-bit quantization give you some control over speed and memory on smaller GPUs. That makes it a practical pick for podcast production, dialogue demos, or character narration where you want natural-sounding multi-speaker audio without leaving ComfyUI.