First I tried to train with Flux.1 Dev using 2048x2048 pixel images, but was too much, as the base model (Flux.1 Dev) was trained using 1024x1024 images. Therefore, my trainign failed, but now I've trained this LoRA using my images without upscaling them x2 (so they are 1024x1024). And now should work.
The images on the testing prompts are not as similar to the initial images much as they are between them. However, I think it is a good LoRA.
