Sign In

anima-sharpen v1.2

23

Updated: May 28, 2026

stylesharpen

Download

1 variant available

SafeTensor

87.6 MB

Verified:

Type

LoRA

Stats

281

9

11

Reviews

Published

May 28, 2026

Base Model

Anima

Hash

AutoV2
DD5889F802
default creator card background decoration
OMCCOM's Avatar

OMCCOM

The Anima Model is licensed by CircleStone Labs LLC. Copyright CircleStone Labs LLC. IN NO EVENT SHALL CIRCLESTONE LABS LLC BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.

Built on NVIDIA Cosmos

anima-sharpen v1.2

This is a LoRA slider for adjusting sharpness.

This time, to match Anima Base 1.0, I changed the training resolution to 1024x1024 and retrained the model.

Since I disabled TextEncoder training, it may work with applications other than ComfuUI.

For sharper images, a LoRA weight of around +0.45 is recommended.

For blurred images, a LoRA weight of around -0.45 is recommended.

シャープネスを調整するスライダーLoRAです。

今回はAnima Base 1.0に合せて、学習解像度を1024x1024に変更し再学習しました。

TextEncoder学習を切ったのでComfuUI以外でも動作するかもしれません。

シャープにしたい場合、LoRAのウェイトは+0.45位が良いと思います。

ボカしたい場合、LoRAのウェイトは-0.45位が良いと思います。

Notes:

I investigated the issue with the sharpness direction in v1.1, as it wasn't satisfactory.

As a result, I determined that the problem wasn't due to insufficient steps during overfitting, but rather because the model was overfitting and breaking down due to excessive steps.

Therefore, I divided the epoch into smaller segments, performed visual selection, and merged the LoRA.

v1.2 is improved, so please try it out.

v1.1のシャープネス方向が良好でないので調査しました。

結果、過学習時のステップ数が不足していないのではなく、ステップが進み過ぎ過学習モデルが破綻している事に起因していると判断しました。

そこでエポックを細かく区切り目視による選別を行いLoRAをマージしました。

v1.2は改善しているのでお試し下さい。