Microsoft VibeVoice TTS
Run Microsoft's VibeVoice text-to-speech locally on Windows using either Gradio or ComfyUI, with custom voice-cloning support. The original Microsoft repository was taken down, so this guide uses a maintained backup fork.
✓ 100% free — download the workflow and run it locally.Requirements
- OS: Windows
- GPU: NVIDIA (CUDA-compatible)
- Python: 3.12 (Miniconda or direct installer)
- Also needed: Git, FFmpeg, NVIDIA CUDA toolkit + latest drivers
- Storage: enough for the 1.5B or 7B model weights
Model downloads
- VibeVoice 1.5B https://huggingface.co/microsoft/VibeVoice-1.5B/tree/main
- VibeVoice 7B (Large) https://huggingface.co/aoi-ot/VibeVoice-Large/tree/main
Custom nodes
- VibeVoice Fork (backup of Microsoft repo) https://github.com/TheLocalLab/VibeVoice-Fork
- ComfyUI-VibeVoice node https://github.com/wildminder/ComfyUI-VibeVoice
- ComfyUI Manager https://github.com/ltdrdata/ComfyUI-Manager
Setup steps
- Gradio: clone the fork — git clone https://github.com/TheLocalLab/VibeVoice-Fork.git
- Gradio: install it — pip install -e .
- Install PyTorch (CUDA 12.8) — pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128
- Download a Flash Attention wheel from github.com/kingbri1/flash-attention/releases and install it — pip install <wheel_filename>
- Install Triton — pip install -U "triton-windows<3.5"
- ComfyUI: clone both custom nodes into ComfyUI/custom_nodes/ and install their requirements.
- Download model weights to ComfyUI/models/tts/VibeVoice/, launch ComfyUI, and load the example workflow below.
Notes & tips
- Set the CFG slider to ~1.30–1.35 for the best results.
- Add voice samples (WAV/MP3) to VibeVoice-Fork/demo/voices/ for Gradio; ComfyUI accepts WAV or MP3 input.
- Prefer a one-click install? A ready-made VibeVoice installer is available.
Want the 1-click version?
Skip the manual model downloads and node installs — get a ready-to-run installer, or unlock all 75+ with Local Lab Pro.