Running the Flux.dev image generation model locally
Quick notes on steps I'm taking to run Flux model YOU WILL PROBABLY WANT TO SKIP ALL THIS and just use ComfyUI! Skip down if you've randomly found my wiki :
I created a venv, pip installed:
pip install --upgrade pip pip install git+https://github.com/huggingface/diffusers.git pip install torch pip install transformers pip install -U "huggingface_hub[cli]"
you need to make a hugging face username and agree to their terms
create a key and login with huggingface-cli login
pip install sentencepiece pip install protobuf pip install accelerate
I couldn't run the examples and found a ComfyUI pipeline that might work. I had to download a file and then move it out of the cache using the hugging-face cli:
huggingface-cli download black-forest-labs/FLUX.1-dev flux1-dev.sft
the other files I was able to just use wget.
more pip install for ComfyUI:
pip install -r requirements.txt
since I'm running on another box with an good card, I have to run ComfyUI with --listen (listen defaults to 0.0.0.0 without arg, or 127.0.0.0 if listen switch isn't provided)
navigate to http://$IP:8188 and drag the image over from the site linked above to setup the workflow
I had to fix the symlink to the flux1-dev.sft model because huggingface-cli used a relative path.
After that it worked!
