comfy-pack

comfy-pack, maintained by the BentoML team, addresses the reproducibility problems that plague shared ComfyUI workflows: missing custom nodes, mismatched model files, and Python dependency conflicts. It packs a working workflow into a .cpack.zip artifact that records the Python package set, the exact git revisions of ComfyUI and every custom node, and hashes of the models used. Unpacking on another machine rebuilds the environment automatically, cloning repositories at the pinned revisions and locating models from registries such as Hugging Face and Civitai, with symlinking to avoid duplicate downloads. Beyond sharing, workflows can be deployed as REST APIs with configurable parameters through BentoML, accessible over HTTP or from a Python client. The artifact stores only metadata rather than weights, though unpacking installs third-party node code, so sources still need to be trusted. The open source tool installs through ComfyUI Manager or pip and targets creators and teams shipping ComfyUI pipelines.