Skip to content
Newsostris Releases OpenFLUX.1: A Commercially Usable De-distilled Version of FLUX.1-schnell

ostris Releases OpenFLUX.1: A Commercially Usable De-distilled Version of FLUX.1-schnell

ostris has recently released a new model called OpenFLUX.1 on Hugging Face. This model is a special version of FLUX.1-schnell that has undergone de-distillation, offering users more flexibility and possibilities. Notably, unlike Flux dev which cannot be used commercially, OpenFLUX.1 retains the commercial license of FLUX.1-schnell, making it significantly more valuable for practical applications.

OpenFLux1.1

Features of OpenFLUX.1

  1. De-distillation: The original FLUX.1-schnell was a distilled model that couldn’t be fine-tuned. OpenFLUX.1 removes the distillation effect, allowing the model to be further fine-tuned and customized.

  2. Normal CFG Values: Unlike the original model, OpenFLUX.1 uses classic Classifier-Free Guidance (CFG). ostris suggests that a CFG value of 3.5 works well.

  3. New Inference Pipeline: Due to the changes in the model structure, OpenFLUX.1 requires a different inference pipeline than the original FLUX.1 schnell and dev models. The new pipeline can be found in the open_flux_pipeline.py file in the project repository.

  4. Open-Source and Commercially Usable: OpenFLUX.1 inherits the Apache 2.0 license from FLUX.1-schnell, meaning it’s not only an open-source model that can be freely used and modified but also available for commercial use. This is in stark contrast to Flux dev, which does not allow commercial usage.

Development Background

In the model card, ostris mentions that developing OpenFLUX.1 required significant computational resources and multiple iterations. Although the current version is considered beta, ostris states that the model has reached a satisfactory level, with most of the distillation effects removed.

Commercial Value

The release of OpenFLUX.1 represents a significant opportunity for businesses and developers seeking a high-quality, customizable AI image generation model with commercial licensing. It offers the powerful capabilities of FLUX.1-schnell while adding the possibility of fine-tuning, which is particularly important for commercial applications.

Future Prospects

ostris plans to continue improving the OpenFLUX.1 model. Users can start using and fine-tuning the current version but should be aware that the model may receive updates in the future.

Conclusion

The release of OpenFLUX.1 brings new possibilities to the field of AI image generation. By removing the distillation effect while maintaining open-source and commercial licensing, this model provides researchers, developers, and businesses with a powerful and flexible tool for further exploration, innovation, and commercial applications.

Source: Hugging Face - ostris/OpenFLUX.1