It seems like a long time ago when AI images were full of nightmarishly distorted renderings and unnameable mixtures, but I still like Midjourney v3, which appeared around the end of that era and was unique in its own way.Regarding this LoRA, it is basically not user-friendly (especially the SD 1.5 version). Sometimes people are not drawn, or multiple triggers may need to be emphasized in order to resemble the style.For the Pony version, it is recommended to use it with derivatives other than the original PonyDiffusionV6.AI画像といえば、描画が何もかも歪んでいたり名状しがたい物体ばかりだった時代もまるで昔の事みたいですが、Midjourney v3はそんな時代の最後らへんに現れて異彩を放っていた気がします。このLoRAは基本的に使い勝手が悪く(特にSD1.5版)、時々人物が出せなかったり、スタイルを似せるためにはいくつかトリガー的なタグを強調する必要があるかもしれません。Pony版については、オリジナルのPonyDiffusionV6以外の派生モデルでの使用をお勧めします。

MidjourneyV3Artifacts

text encoder lr:0.125, unet lr:1, prodigy (not reflected in metadata)

mjv3_pony-2.safetensors

Files with SHA256: c81c5c7318fe136cf6fcad25ca92be214199f5655cf21e7fea33422b67679ce0