finds.dev← search

// the find

camenduru/comfyui-colab

★ 545 · Jupyter Notebook · Unlicense · updated Aug 2024

comfyui colabs templates new nodes

A collection of pre-wired Google Colab notebooks for running ComfyUI — the node-graph Stable Diffusion frontend — against various SD 1.x and 2.x era checkpoints. The repo is organized into lite/stable/nightly tiers and covers over 100 model configurations. It's aimed at people who want to try ComfyUI's workflow editor without owning a local GPU.

The lite/stable/nightly split is a practical design decision: lite targets Colab's memory-constrained free tier, stable pins extension versions for reproducibility, nightly tracks ComfyUI HEAD for those who want the latest nodes. The consistent structure across 100+ notebooks means setup is predictable — one notebook is effectively all of them. Template JSON files are included alongside the notebooks so you can import pre-built ComfyUI graphs directly without recreating them from scratch. Separating `_colab.ipynb` (environment setup) from `_run.ipynb` (execution) is a sensible split for iterating without re-downloading 4GB checkpoints every session.

The repo was last touched August 2024 and the models it covers are almost entirely from the SD 1.x/2.x era — no SDXL, no Flux, no SD3, nothing from the last two years of actual development. If you're looking to run anything modern, this won't help. Google Colab's free tier has progressively throttled GPU access since these notebooks were written, and nothing in the repo acknowledges this; people will hit session limits and have no idea why. Model download URLs point directly to CivitAI and HuggingFace repos that can disappear without warning — when they do, the notebooks fail silently mid-run with a download error and no fallback. There's also no documentation on ComfyUI itself: you get a running instance but zero guidance on what the nodes do, how to wire your own graph, or what parameters to adjust.

View on GitHub →

// want more like this?

We dig through GitHub every week and send a few repos picked for what you actually care about — each with an honest take like this one.

Get finds in your inbox → Search again →