This is a Flux Dev retrain of my SDXL model Frosting Lane Redux. I used Ostris' AI Toolkit trainer.

Special thanks to @ostris for creating an excellent codebase for model training. https://github.com/ostris/ai-toolkit/tree/main

This is the first test of training a model, and all licensing conditions for the Flux Dev model apply. https://huggingface.co/black-forest-labs/FLUX.1-dev. It's a great model!

A huge thank you to Datacrunch.io for generously providing their resources for training and support. I highly recommend them.

The purpose of this model is to enhance understanding of ??????? and improve prompt comprehension. The differences are subtle but may help you correct some mistakes. I tested it on approximately 300 generations, and in 90% of the cases, the results were better in terms of ??????? and realism. It might distort your text, so don't be afraid to experiment and adjust the model's weight. I tested it using seed 0 with a model weight of 1. The model was trained on approximately 5,000 images. If the test results are mostly positive, I will train it on a larger dataset. I welcome any feedback in the comments—feel free to share any errors or shortcomings of the model — this is just an experiment.

(the first image features the use of a LoRA, while the second image does not.)