My apologies, that v3 was dogshit overtrained. this is much less epochs and much much better. tf was i thinking. V4 probably coming, this still hasnt captured style yetEdit V3: Improved captions, 10 more images, more epochsEdit V2: Just better than V1 for both style and flexibilityStyle for krea2. 800 for minor style change but better anatomy, 1300 for better style. steps on Onetrainer batch size 2. A more chibi(not really) expressive style look. first two images are comparision between lora and default. i use kroma lora for all images. didnt want the style to affect composition a lot so pretty happy with it

Files with SHA256: 617e166227bf2ddb93ee123078780a7de9c13f21ea6a3c9d0d51c233e42df10c