It's my first experience in creating LORA. Model trained on screenshots from anime on AnythingV3. I highly recomend to use upscaler for good results. I used R-ESRGAN 4x+. Also I used DDIM with 40 steps and 11-12 CFG. It should good works with AnythingV3, wounddedMix, wildSMix_v1Alloround and similar. With AOM the images are very dim (and I could not solve it). Can be corrected in Photoshop by increasing the contrast. Sample prompts:<lora:witch_maria:0.75>, witch_maria, 1girl, jewelry, blonde hair, solo, blue eyes, earrings, black dress, coat. Feel free to ask questions here or in discord VanDerGraf#7304 |