arako o | あら湖 | noob v-pred test

Training procedure: https://rentry.org/lora-is-not-a-finetune

Resumed from: https://civitai.com/models/833294?modelVersionId=962003

+ https://github.com/kohya-ss/sd-scripts/pull/1715

+ scale_v_pred_loss_like_noise_pred

+ Quasi-Register Tokens

Dataset (thanks to anon): https://files.catbox.moe/pbu0go.rar