Text-to-Image

Flux-Mini Cross Post & Conversion

#3
by dasilva333 - opened

Hey guys great work, excellent model, delivering outstanding results, you guys should try it with the SharkSampler to see what better output it can produce than standard KSampler.

Anyways I posted the model here giving attribution and linking to your project appropriately: https://civitai.com/models/955242

I also created a script that converts the model to a format that's compatible with the Load Checkpoint node in ComfyUI so that users aren't so confused how to use it.

This is stated in the model's page but I'll say it here again in case anyone comes across it that needs to know:

  • Load Checkpoint Node will not work with the model
  • Load Diffuusion Model node must be used instead
  • That means placing the safetensors file in the models/diffusion_models directory instead of checkpoints
  • After that you can set it up how you always do for any other flux setup, getting the CLIP and VAE from DualClipLoader and Load VAE

Expect to see an update on the Civit page where I will post the .safetensors version that IS compatible with Load Checkpoint.
I also quantsized and posted the model reducing the file size even further from 5gb to 3.4gb using Q8 which in theory should only be a 0.1% difference from the original model.

Sign up or log in to comment