ostris / openflux.1

OpenFLUX.1 (Beta v0.1.0), is a fine tune of the FLUX.1-Schnell model with the distillation trained out of it

  • Public
  • 276 runs
  • GitHub
  • Weights
  • License
Run with an API
Playground API Examples README Versions

Examples

View more

Run time and cost

This model costs approximately $0.10 to run on Replicate, or 10 runs per $1, but this varies depending on your inputs. It is also open source and you can run it on your own computer with Docker.

This model runs on Nvidia A100 (80GB) GPU hardware. Predictions typically complete within 74 seconds. The predict time for this model varies significantly based on the inputs.

Readme

Beta Version v0.1.0

After numerous iterations and spending way too much of my own money on compute to train this, I think it is finally at the point I am happy to consider it a beta. I am still going to continue to train it, but the distillation has been mostly trained out of it at this point. So phase 1 is complete. Feel free to use it and fine tune it, but be aware that I will likely continue to update it.

What is this?

This is a fine tune of the FLUX.1-schnell model that has had the distillation trained out of it. Flux Schnell is licensed Apache 2.0, but it is a distilled model, meaning you cannot fine-tune it. However, it is an amazing model that can generate amazing images in 1-4 steps. This is an attempt to remove the distillation to create an open source, permissivle licensed model that can be fine tuned.

How to Use

Since the distillation has been fine tuned out of the model, it uses classic CFG. Since it requires CFG, it will require a different pipeline than the original FLUX.1 schnell and dev models. This pipeline can be found in open_flux_pipeline.py in this repo. I will be adding example code in the next few days, but for now, a cfg of 3.5 seems to work well.