anima-boldline v1.2

세부 정보

파일 다운로드 (1)

모델 설명

anima-boldline v1.2

This is a LoRA slider similar to sdxl-boldline, which thickens or thins the main lines.

It has been retrained to coincide with the release of Anima Base V1.0. Since it no longer trains the TextEncoder, it should work with Forge Neo as well. The training resolution has also been changed from 512x512 to 1024x1024.

If you want to make the main lines thicker, the LoRA weight should be around +0.25, and if you want to make them thinner, the limit is probably around -0.25. When making them thicker, a value of +0.5 results in a rather flat style, and on the negative side, it seems to alter the expression and make the style more flat.

sdxl-boldlineの様に主線を太くしたり細くしたりするスライダーLoRAです。

Anima Base V1.0のリリースに合せ再学習しました、今回からTextEncoderの学習は行っていないのでForge Neoでも使えると思います。また学習解像度を512x512から1024x1024に変更しています。

主線を太くしたい場合、LoRAのウェイトは+0.25位、主線を細くしたい場合、LoRAのウェイトは-0.25位が限界でしょうか。太くする場合+0.5では大分ベタ付いた画風に変化してしまいます、またマイナス側では表情の変化やフラット傾向になる様子です。

Notes:

Please note that there will be some variation in style and composition; even with the same seed, the composition will not be exactly the same.

Based on the findings from anima-sharpen v1.2, I thought anima-boldline also needed to perform visual selection of overfitted LoRA models, so I tried it and found that it improved the problem of the art style becoming rough, especially with a weight of +0.5, so I updated it. That being said, at ±0.5, there is still a significant impact on composition and faces in addition to the art style.

I've decided to introduce the fox character this time, and for comparison, I'll also include the holo results this time only.

画風や構図に多少なりとも変化が加わります、同一シードであっても全く同一の構図にはならない点に注意して下さい。

anima-sharpen v1.2の知見から、anima-boldlineも目視による過学習LoRAの選別を行う必要があると思い試行した所、特に+0.5のウェイトでの画風が荒くなる問題の改善が見られたので更新しました。とは言え±0.5では画風以外にも構図や顔への影響が大分多いです。

今回からキツネさんに登場してもらう事にしました、比較の為にホロの結果も今回だけは掲載します。

이 모델로 만든 이미지