<h2>Yoshida Hirofumi 吉田ヒロフミ (Chainsaw Man)</h2>Triggers with <code>yoshida hirofumi</code>, weight 1 works best with most models.Used Cetus-mix for example images.It was really a journey to make this LoRA. For such an easy design, it was actually complicated to make it work. It took me 13 tries to get it close to perfection, but well, I think it was worth it.<h2>Usage</h2>General tags: <code>yoshida_csm</code> works fine already, but you can use <code>1boy, black hair, black eyes, mole under mouth, short hair, crossed bangs</code> if necessary.For Yoshida's gakuran, just use <code>gakuran, buttons, black pants</code>, sometimes you need to put <code>green pants</code> in the negative prompt (?).For Yoshida's casual clothes, use <code>official alternate costume, open jacket, white jacket, black pants.</code><h2>V2.0 CHANGELOGS</h2><ul><li>More stability and quality than v1.0.</li><li>Changed the training parameters.</li><li>More flexibility, can easily change clothes.</li></ul><h2>TRAINING PARAMETERS</h2><ul><li>Trained on the old version of kohya-LoRA-dreambooth on Google Colab.</li><li>Trained on a set of 35 images.<ul><li>10 epochs, 2 batch.</li><li>8 repeats</li></ul></li><li>0.0001 LR + 0.0001 unet_LR + 1e-05 text_encoder_LR.</li><li><code>cosine_with_restars</code> scheduler, 0 lr_warmup_steps.</li><li>64.0 net_alpha, 128 net_dim.</li><li>Training resolution 896x896.</li><li>Model: NAI Model (animefull-final-pruned.ckpt), no VAE.</li><li>225 max_token_length.</li><li>fp16 mixed_precision.</li></ul>

Yoshida Hirofumi 吉田ヒロフミ (Chainsaw Man)

<ul><li>Trained on the NAI model, used Kohya-Dreambooth colab.</li><li>4 batch size, 3 epochs.</li><li>1.5e-5 text encoder LR</li><li>1.5e-4 unet LR</li><li>1e-5 optimizer LR</li><li>Training resolution 768x768.</li></ul>

Files with SHA256: bc14475cb801e9b3c725c45f65d34eee5bf4f82b05b2ef870afee35a6f040f7c