>Waifu Diffusion is the name for this project of fine tuning Stable Diffusion on Danbooru images.
>The current model has been fine-tuned with a learning rate of 5.0e-6 for 4 epochs on 56k Danbooru text-image pairs which all have an aesthetic rating greater than 6.0.
>@forasteran: #WaifuDiffusion まとめ (discordより)
>現行 WD 1.3
>- SD1.4 ベース
>- 学習データ 68万枚
>―――
>次 WD 1.3.5: 今週公開予定🆕
>- SD1.5 ベース
>- 学習データ 約600万枚 (1.3比で約8.8倍)
>―――
>その次 WD 1.4: 12月24日公開予定🎅
>- SD2.1 ベース
>- 学習データ 約1500万枚 (1.3比で約22倍)