cog-comfyui

Why we included this project

ComfyUI is great for designing image workflows, but wiring one into your own application is a separate problem. This project wraps ComfyUI in a Cog container so a workflow's API JSON runs over HTTP, and it handles the fiddly parts: downloading model weights, installing custom nodes, and accepting image inputs as URLs, single uploads, or zipped folders. It ships with a curated set of popular weights and custom nodes, and the README shows how to pull LoRAs straight from CivitAI or HuggingFace by URL. It also walks through the trade-offs between running on a shared public model versus a dedicated instance, which matters when you need predictable latency for repeated image-generation calls. If you've got a node graph you want to turn into a repeatable image service, this is a solid starting point.