+I re-trained using official cutout illustraion for similar art style. I used about 40 official character illustrations to train this LoRA. The base model is the animefull-final-pruned model. The example images were generated using a merge model based on AbyssOrnageMix2. For a similar look to the original illustration: add small breasts to the prompt add loli to the negative prompt. My recommended settings are: Sampling method: DPM++ 2M Karras, for fast and high quality. CFG scale: 5 For the rest of the settings, follow the model's recommended values. This LoRA trained to 20epochs. |