FLUX.1 models are released by Black Forrest Labs, which is founded by developers of Stable Diffusion models.
A really short introduction
FLUX.1 models are high quality Diffusion Transformer models. Unlike previous diffusion models:
- It understands better image composition and better anotomy of human body.
- It is quite capable of writting different styles of text into images.
- It is more supirior in terms of aethetics and image quality.
However, that doesn't come without cost. Though the community quickly improves tools around it, there still be drawbacks if you want to try it, namely
- It takes a fairly beef GPU to load all model weights and run it.
- It cannot be used to take images as input to generate modified version, just like Stable Diffusion do.
Using FLUX.1 models
Fine tuning with Dreambooth
Since the announcement of FLUX.1 models, the community put a lot of efforts to get Dreambooth work for FLUX models.
Since the Schell version is step distilled, it can hardly be fine tuned to learn a new style or new person or object.
Generate images with ControlNet
We at PixelsAI provides easy-to-use fine tuning service for FLUX.1 dev models. It enables you to generate hyperrealistic photos of yourself with only 10s of your photos. We're rolling out beta versions to our users.