A ComfyUI node pack around a native engine

ComfyUI-QuantFunc is a ComfyUI plugin for QuantFunc, a native C++ and CUDA diffusion inference engine. The README says it runs quantized text-to-image and image editing models at 2x to 11x speed, with SVDQ offline quantization and Lighting runtime quantization.

The important detail is that this is not a pure Python node pack. It auto-downloads a matching native library, libquantfunc.so on Linux or quantfunc.dll on Windows, from ModelScope on first startup. That makes setup easier, but it also means your trust boundary includes a binary engine and its updater.

Install path

The recommended path is to clone the plugin into ComfyUI’s custom nodes directory:

cd ComfyUI/custom_nodes
git clone https://github.com/QuantFunc/ComfyUI-QuantFunc.git

On first start, the plugin downloads the matching engine binary. If the auto-download is skipped, the README says you can manually place libquantfunc.so under bin/linux/ or quantfunc.dll under bin/windows/.

Hardware and software requirements

The README lists NVIDIA RTX 20 series or newer, 8 GB VRAM, driver 560 or newer, CUDA Runtime 13.0 or newer, cuDNN 9.x, Linux with glibc 2.31 or Windows 10/11, and Python 3.9 or newer through ComfyUI’s embedded Python. It also documents CUDA 12 and CUDA 13 runtime package options for Linux.

This is a narrow target. If you are on macOS, AMD GPU, old NVIDIA cards, or a locked-down environment that cannot load downloaded binaries, this is probably not the right plugin.

What changed in current README

The README highlights plugin 0.0.02 and engine 0.0.07. Notable additions include v2 loaders, universal format adapters for diffusers, BFL, Flux, nunchaku SVDQ, bundled checkpoint and HF layouts, inpainting, QwenImage, QwenImage-Edit, Flux.2 Klein, worker-process cleanup, zip-slip guard, and IPC bound checks.

The performance claim is attractive, but verify it on your exact model, GPU, driver, and workflow. Quantized diffusion acceleration is sensitive to hardware and model format.

For other AI media tooling on this site, see harry0703/MoneyPrinterTurbo, OpenCut-app/OpenCut, and heygen-com/hyperframes.

FAQ

Is ComfyUI-QuantFunc pure Python? No. It is a ComfyUI plugin that loads a native QuantFunc engine binary.

Does it support Windows? The README lists Windows 10/11 and a quantfunc.dll path.

What GPU do I need? The README lists NVIDIA RTX 20 series or newer and at least 8 GB VRAM.

What should I check before installing? Check your CUDA, driver, cuDNN, GPU generation, and whether you are comfortable with auto-downloaded native binaries.