copycat-anima - 20260216

copycat-anima

CHECKPOINT
Reprint


Updated:

copycat-anima by miaomiao on Tensor.Art
copycat-anima by miaomiao on Tensor.Art
copycat-anima by miaomiao on Tensor.Art
copycat-anima by miaomiao on Tensor.Art
copycat-anima by miaomiao on Tensor.Art
copycat-anima by miaomiao on Tensor.Art
copycat-anima by miaomiao on Tensor.Art

creator: calculater

ver.20260425/20260329/20260302

Quality tag:May or may not be used. / 使っても良いし、使わなくても良いかも。

 

ver.20260425

 anima-preview3をベースに再調整。

 Readjusted based on anima-preview3.

 

ver.20260329

 anima-preview2をベースにしたファインチューンモデル。

 主に背景に関するデータセットを変更。

 ver.20260302と同様にクオリティタグは必須ではありません。

 A fine-tuned model based on anima-preview2.

 Mainly updated the dataset related to the background.

 As with ver.20260302, the quality tag is not required.

 

 

ver.20260302

 anima-previewをベースにしたファインチューンモデル。

 ポジティブ/ネガティブプロンプトのクオリティタグは必須ではありません。(サンプル画像は全てクオリティタグを使用せず出力しているものです)

 画風はアニメ塗りに寄せています。

 A fine-tuned model based on anima-preview.

 Positive/negative prompt quality tags are not required. (All sample images were generated without using quality tags.)

 The art style leans toward anime-style coloring.

 

追記

 ver.20260216と同様に、モデル調整後に最終的にベースモデルであるanima-previewをマージさせています。(ウェイト約0.4でマージ)

 追加学習を行ったが故に構図の柔軟性が損なわれてしまうのを抑えるのが狙いですが、ベースモデルとのマージはその分だけ未学習の状態に戻すということなので、やりすぎると追加学習の効果が薄まる/消えることになります。

 そのため、今回は視覚的に判別しやすいアニメ塗りを学習させ、(クオリティタグなしでも)アニメ塗りの画風が崩れない範囲でベースモデルのマージ割合を引き上げる、という手法を取りました。

 

Additional note

 Similar to ver.20260216, we ultimately merged the base model anima-preview after model adjustment. (Merged with a weight of approximately 0.4)

The goal is to suppress the loss of compositional flexibility that occurs due to additional training. However, merging with the base model effectively reverts the model to an untrained state to that extent. Overdoing this diminishes or eliminates the effects of the additional training.

 Therefore, this time we employed a technique where we trained the model to produce visually distinguishable anime-style coloring, then increased the base model's merge ratio within the range where the anime-style coloring remains intact (even without a quality tag).

 

ver.20260216

 ・Negative Prompt

  worst quality,low quality,score_1,score_2,score_3

 

 animaをベースに、AI生成画像を元に作成したLoraをマージしたモデルです。

 背景用画像ないしnsfw画像で作成したloraで調整したモデルを作成。そのモデルにベースモデルであるanima-previewをウェイト約0.5でマージし直したもの。

 A model created by merging Lora with AI-generated images based on anima.

 Created a model adjusted with LORA using background images or NSFW images, then re-merged it with the base model anima-preview at a weight of approximately 0.5.

 

ver.20260209

 やや童顔の出力強め。

 The output has a slightly stronger baby-faced appearance.

Version Detail

Anima
based on anima(https://huggingface.co/circlestone-labs/Anima). Special thanks to the authors of the models anima.

Project Permissions

Model reprinted from : https://civitai.red/models/2377376

Reprinted models are for communication and learning purposes only, not for commercial use. Original authors can contact us to transfer the models through our Discord channel --- #claim-models.

Related Posts