Sign In

Color Grade 01

Updated: May 31, 2026

style

Download

1 variant available

SafeTensor

43.85 MB

Verified:

Type

LoRA

Stats

198

Reviews

Published

May 29, 2026

Base Model

Anima

Hash

AutoV2
6517314C13
default creator card background decoration
Bronze Style Badge
SI

siouni

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

Color Grade 01

Color Grade 01 は、暖色・セピア寄りになりやすい出力を、ややクールでクリアな方向へ調整する色調LoRAです。
赤みや黄みを少し抑え、白や青をすっきり見せることで、軽い透明感や抜け感が出やすくなります。
絵柄そのものを大きく変えるのではなく、スタイルを保ちながら色味と空気感を整えることを目的としています。
なお、調整は完全に中立ではなく、やや青・緑寄りの主観的な色味になります。

Anima base v1をベースにADDifT学習を使用して作成しています。

こちらのsd-scriptsフォーク版を使用しました。
これがなければこのLoRAは作れませんでした。
https://github.com/tukisuwa/sd-scripts

ADDifT学習は、ざっくり言えば画像の差分でLoRAを作成します。
Anima base v1そのままの出力と、別途作成したスタイルLoRA 6種類の出力の計7枚を手動で色調調整し、その生成プロンプト、修正前の画像、修正後の画像をデータセットとしてADDifT学習しました。

(This English description was translated with the assistance of ChatGPT.)

Color Grade 01

Color Grade 01 is a color-grading LoRA designed to shift outputs that tend to look warm or sepia-toned toward a slightly cooler and clearer direction.
By slightly reducing red and yellow tones and making whites and blues appear cleaner, it tends to create a light sense of clarity and openness.
Rather than significantly changing the illustration style itself, this LoRA is intended to adjust the color tone and overall atmosphere while preserving the original style as much as possible.
Please note that this is not a perfectly neutral adjustment. It reflects a subjective color preference with a slight blue-green bias.

This LoRA was created for Anima base v1 using ADDifT training.

I used the following fork of sd-scripts:
https://github.com/tukisuwa/sd-scripts
This LoRA would not have been possible without it.

Roughly speaking, ADDifT training creates a LoRA from differences between images.
I manually color-graded a total of seven images: one output from the unmodified Anima base v1 model and six outputs generated with my separately created style LoRAs. I then trained this LoRA with ADDifT using the generation prompts, the original images before color grading, and the corresponding images after color grading as the dataset.