image
imagewidth (px)
384
2.91k
label
class label
7 classes
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50
0aesthetic_50

私がnijijourney v5で生成した画像。自由に使えます。(けど詐欺とか犯罪につかうのはやめてね)

おすすめの使い方としては、とりあえず中の画像を見てみて好きなものだけ選んで使うとよいと思います。

全体の注意点として、必ずしもすべての画像にキャプションが付属してるとは限らないのと、キャプションがついていてもそのまま使うと問題が発生する可能性があるのであまり信用しないでください。

また人為的なミスにより、4分割されずに結合している画像があったり、過度に分割されている画像があったりするので注意してください。

vol1

だいたい2000枚くらいで、多分全部デフォルトスタイルのものです。

解答すると中にLAION Aesthetic v2のスコアでいくつかのフォルダに分類されています。aesthetic_50 ならスコア0.5以上のものです。not_aesthetic は0.5未満のものです。

ただし、exceptional フォルダはチェリーピックした画像が入っており、aesthetic_xx の中のものと重複します。exclude フォルダは、主観で奇妙なものを除いたものです。

aesthetic_xxexceptional にはキャプション(BLIP2、Tagger)ファイルがついていますが、いろいろ変な調整しているのでおそらく各自でキャプションをつけ直したほうがいいと思います。

vol2

だいたい1200枚くらいで、複数のスタイルモードで生成したものが含まれます。

手動でスタイルごとにフォルダを分けています。

defaultcuteexpressivescenic はそれぞれのスタイルっぽい画像で分類していますが、たまに分類を間違えています(ごめん)。

clearrough painting は、個人的に近いと思ったスタイルの画像を入れていて、4つのスタイルの画像とは重複しません。

default cute expressive scenic clear rough painting

vol3

450枚くらいです。キャプションは一切ついていません。

雰囲気で分類しています。

  • background: 背景のみで人物が写っていないもの
  • comic: 白黒だったり漫画風なもの(百合が多いので微妙に注意)
  • ink painting: 水墨画・水彩っぽいもの
  • scenic: scenic っぽい画像で、人物が写っているものも含む。background の画像と一部重複する。

含まれる画像の例

background comic ink painting scenic

vol4

現在は48枚です。

  • minimalist: 非常にシンプルな感じの画風の画像

含まれる画像の例

minimalist
Downloads last month
2
Edit dataset card

Models trained or fine-tuned on p1atdev/niji-v5