This is a Cog training model that creates LoRA-based fine-tunes for the FLUX.1 family of image generation models.
It's live at replicate.com/ostris/flux-dev-lora-trainer.
It also includes code for running inference with a fine-tuned model.
- Automatic image captioning during training
- Image generation using the LoRA (inference)
- Optionally uploads fine-tuned weights to Hugging Face after training
- Automated test suite with cog-safe-push for continuous deployment
- Weights and biases integration
If you're looking to create your own fine-tuned model on Replicate, you don't need to do anything with this codebase.
Check out these guides to get started:
👉 Fine-tune Flux to create images of yourself
If you're here to help improve the trainer that Replicate uses to fine-tune Flux models, you've come to the right place.
Check out the contributing guide to get started.
This project is based on the ai-toolkit project, which was created by @ostris. ❤️
The code in this repository is licensed under the Apache-2.0 License.
The ai-toolkit project is licensed under the MIT License.
Flux Dev falls under the FLUX.1 [dev] Non-Commercial License.
FLUX.1 [dev] fine-tuned weights and their outputs are non-commercial by default, but can be used commercially when running on Replicate.
Flux Schnell falls under the Apache-2.0 License.
