// the find
kijai/ComfyUI-FramePackWrapper
A ComfyUI custom node wrapper for lllyasviel's FramePack, which does image-to-video generation using HunyuanVideo. It lets ComfyUI users drive FramePack through the node graph interface instead of the original Gradio demo. Aimed at people already deep in the ComfyUI ecosystem who want video generation without leaving their existing workflows.
Ships fp8 quantized weights that actually run on consumer VRAM budgets. Reuses the Comfy-Org repackaged model files so you're not downloading yet another copy of HunyuanVideo's encoders if you already have them. Includes a working example workflow JSON so you're not starting from a blank canvas. The memory management module (diffusers_helper/memory.py) looks like it was pulled and adapted from lllyasviel's original, meaning the VRAM offloading tricks carry over.
README says 'WORK IN PROGRESS' in large text and the repo has barely any documentation beyond model download links — you're on your own if something breaks. The codebase is essentially a lightly modified copy of upstream FramePack internals vendored into diffusers_helper/, which means tracking upstream fixes is a manual, error-prone process. Last pushed January 2026 but FramePack itself is still moving fast, so drift is likely. No tests, no CI, no version pinning in requirements.txt — the usual ComfyUI custom node situation where it works until a transitive dependency updates.