New version (Wan and Flux2.0) have a completely reworked dataset and all dalle3 images have been replaced.


Old version was Trained on 27 dalle3 generated images. Works great for text2img, but also I find it very nice with controlnet and img2img.