.png)
3 months ago
2
This is the official repository for FLUX.1 Krea [dev] (AKA flux-krea).
The code in this repository and the weights hosted on Huggingface are the open version of Krea 1, our first image model trained in collaboration with Black Forest Labs to offer superior aesthetic control and image quality.
The repository contains inference code and a Jupyter Notebook to run the model; you can download the weights and inspect the model card here.
Check inference.ipynb for a full example. It may take a few minutes to download the model weights on your first attempt.
Recommended inference settings
-
Resolution - between 1024 and 1280 pixels.
-
Number of inference steps - between 28 - 32 steps
-
CFG Guidance - between 3.5 - 5.0
Krea 1 was created in as a research collaboration between Krea and Black Forest Labs.
FLUX.1 Krea [dev] is a 12B param. rectified-flow model distilled from Krea 1. This model is a CFG-distilled model and fully compatible with the FLUX.1 [dev] architecture.
In a nutshell, we ran a large-scale post-training of the pre-trained weights provided by Black Forest Labs.
For more details on the development of this model, read our technical blog post.
We would like to thank the Black Forest Labs team for providing the base model weights. None of this would be possible without their contribution. The post-training work would not be possible without the hard work of our data, infrastructure, and product team who put together a solid foundation for our post-training pipelines.
If you are interested in building large-scale image/video/3D/world models, or the engineering and data infrastructure around it...
