ComfyUI LTS
author, 2023–2026
A long-term-support fork of ComfyUI installs as a standard Python package and scales across container clusters. It is used in production by Scopely, Livepeer, and Nunchaku Tech.
ComfyUI is a node editor used to run image and video generation models. Standard installations require cloning a code repository and running a launcher script, a process designed for individual workstations that is difficult to manage across server infrastructure. This fork installs with pip or uv like a standard Python package, allowing programmers to import comfy and run it as a library inside an application without starting a web server.
The software automatically downloads missing models from Hugging Face or CivitAI. Maintained container images support hardware from NVIDIA and AMD, while RabbitMQ distributes inference workloads across multiple worker systems and frontends. The architecture supports large language models and multi-modal workflows, and an automated test suite validates software upgrades. Custom nodes from the wider ecosystem remain compatible, distributed through a package index and installed with uv instead of manual repository clones.
The software is used in production by the game studio Scopely, by Livepeer, and by Nunchaku Tech, and in work at Ferrero, Hyundai and Nike.