V1 最初のバージョン。約1200枚の画像を用いて、バッチサイズ4で11300kステップ学習。
プロンプトの後ろの数値は、例えば「short_hair:0.781」のように、そのプロンプトが使用された画像の割合を示しています。これは、プロンプトが78%の画像で使用されたことを意味します。
この指標は、どのプロンプトが強い反応を引き起こすかを把握するのに役立つと思います。
(breasts:1.0), (brown hair:0.881), (large breasts:0.843), (1girl:0.842), (1boy:0.785), (short hair:0.781), (glasses:0.77), (jewelry:0.655), (shirt:0.626), (indoors:0.591), (black hair:0.573), (blonde hair:0.498), (cleavage:0.442), (realistic:0.38), (long hair:0.372), (blue eyes:0.366), (pants:0.347), (looking at another:0.347), (earrings:0.344), (sitting:0.333), (necklace:0.332), (multiple girls:0.314), (bracelet:0.312), (navel:0.289), (facial hair:0.288), (multiple boys:0.278), (underwear:0.247), (couch:0.245), (red shirt:0.241), (solo:0.233), (2girls:0.229), (tattoo:0.222), (hetero:0.22), (dress:0.212), (twintails:0.211), (choker:0.208), (2boys:0.208), (makeup:0.196), (table:0.195), (nipples:0.187), (male focus:0.184), (red hair:0.18), (green eyes:0.173), (belt:0.171), (ponytail:0.171), (bra:0.166), (chair:0.165), (vest:0.163), (mole:0.16), (midriff:0.158), (black-framed eyewear:0.154), (wristwatch:0.146), (nude:0.145), (watch:0.144), (arm tattoo:0.14), (cup:0.13), (crop top:0.125), (panties:0.12), (black pants:0.118), (white shirt:0.116), (standing:0.113), (formal:0.112), (piercing:0.111), (looking at viewer:0.111), (denim:0.11), (holding:0.109), (mole on breast:0.109), (gloves:0.108), (upper body:0.108), (suit:0.108), (shorts:0.105), (jeans:0.105), (lips:0.103), (nail polish:0.101)
