v1: Shirow Masamune style for Anima-preview (trained on v2). I recommend res_multisteps sampler with lora at 08-1.4 strength (1.2 is usually the sweet spot) Number of steps will depend on the model, personally I like <a target="_blank" rel="ugc" href="https://civarchive.com/models/2426265/animaika">Animalka</a> v2.5 at 16 steps CFG 1. With base model I use 40 steps CFG 3.4trigger : @shirow masamune v3: Trained on Anima-Base v1. This time the training is based on the correct trigger : '@shirou masamune', thanks to <a target="_blank" rel="ugc" href="https://civarchive.com/user/ytkuro25145">ytkuro25145 </a>for this hint ! The LoRA works essentially as a style enhancer, see the comparisons in the samples Have fun !

Files with SHA256: f18c7772136f6165051a8177e2335e7a46219be533ad2a92dd03be226b665655