This LoRA is in pre alpha stage. It will be a good flux/hidream/chroma LoRA but i need time to do my best. Many checkpoints there are overtrained/broken. I will try to fix that. I am currently experimenting on how to deal with distillation and other hyperparameters. It looks very problematic, even dev2pro is not a solution here. For now it seem that I need to use far bigger rank than 16 or 32. You can find dataset for this lora here: https://huggingface.co/datasets/GeorgyGUF/adidas tracksuit with three stripes part1 I will improve this dataset too. You can follow me for updates, I will update this repo everyday and I think it will be renamed too. I also learn how to make lora for SDXL and i put my checkpoints there too. I need it to understand which dataset i need without having to deal with distillation. Trying to create best synthetic dataset and to check if it is better then dataset of real images.
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy