// the find
camenduru/flux-jupyter
A collection of Jupyter notebooks for running Black Forest Labs' FLUX.1 image generation models on Colab and Kaggle, using fp8 quantization to fit the models into free-tier GPU memory. Aimed at people who want to try FLUX without paying for compute or wrestling with local setup. This is a quickstart pad, not a library.
fp8 quantization is the right call for Colab's T4 — without it you can't run FLUX.1-dev at all on free hardware. Covers the main variants (dev, schnell, image-to-image, LoRA + Gradio) so you're not hunting across multiple repos. One-click Colab badges mean zero setup friction for the target audience. Kaggle alternative is a nice touch given Colab's increasingly aggressive session limits.
Marked WIP and last pushed October 2024 — FLUX has moved fast since then and these notebooks may already be stale against current diffusers/transformers versions. There's essentially no code here that camenduru wrote; it's thin glue around black-forest-labs/flux and xlabs-ai/x-flux, so if those repos break an API, this silently breaks too. No pinned dependency versions anywhere, which means a notebook that ran last month may not run today. Five notebooks is also a pretty small surface area for 483 stars — the stars are mostly riding FLUX's hype.