Pre-baked ComfyUI for Windows: 40+ nodes, zero compilation tears

What it does This is a curated, portable Windows distribution of ComfyUI that ships with over 40 popular custom nodes and 300+ Python packages pre-installed. It bundles performance libraries like SageAttention and FlashAttention, plus natively compiled binaries such as insightface and dlib, so you don’t need a C++ build environment to get started. You bring your own Stable Diffusion models; the package handles the glue.

The interesting bit The author rewrote ComfyUI’s official packaging script to solve a specific Python Embedded headache: wrapper scripts break when moved because they hardcode absolute paths. The build pipeline is also fully forkable on GitHub, so you can roll your own all-in-one package without configuring CI/CD.

Key highlights

• 40+ custom nodes pre-installed, covering ControlNet, IPAdapter, AnimateDiff, and more
• 300+ Python packages with tricky native dependencies already compiled
• Redirects HuggingFace and PyTorch caches into the program folder to keep everything self-contained
• Supports external model directories via extra_model_paths.yaml for multi-install setups
• Fork the repo and run the GitHub Action to build your own variant in roughly 20–40 minutes

Caveats

• Windows and NVIDIA-only (2018 GPUs or newer); no AMD or macOS support
• First launch may still download some model files, which can freeze on slow connections
• The README warns that “Update All” in ComfyUI-Manager can trigger Python package conflicts

Verdict Worth grabbing if you want to skip the usual Windows dependency ritual and start experimenting with ComfyUI workflows immediately. Skip it if you’re on Linux, macOS, or prefer to curate your own minimal node setup.