Tsundere Service SDXL

A merge of a finetune of Pony and AutismMix Confetti

For the end of the year 2024, I wanted to try my hand on fine tuning SDXL so I researched kohya_ss, rented an RTX 4090 on runpod.io and with some trial and error, the output is something? Some constructive feedback would be helpful.

1. Gathered images 200 images of 50 anime Tsundere characters.

  'agk mine s1',
  'aki adagaki s2',
  'aria holmes kanzaki s1',
  'asuka langley soryu rebuild',
  'azusa nakano s2',
  'chitoge kirisaki s2',
  'chizuru ichinose s1',
  'dia kurosawa s2',
  'eri sawachika s2',
  'erina nakiri s1',
  'eriri spencer sawamura s2',
  'eris boreas greyrat s1',
  'haruhi suzumiya s1',
  'hitagi senjougahara anime',
  'kagami hiiragi s1',
  'kagome higurashi anime',
  'kaguya shinomiya s3',
  'kaho hinata s1',
  'kallen stadtfeld s1',
  'kana arima s2',
  'karane inda s1',
  'kei karuizawa s2',
  'kirino kousaka s2',
  'kotori itsuka s2',
  'krul tepes s2',
  'kurisu makise s1',
  'kyouko sakura s1',
  'louise francoise le blanc de la valliere s3',
  'lucy heartfilia 100yq',
  'lum s1',
  'mafuyu kirisu s2',
  'mai kawakami s1',
  'mai sakurajima s1',
  'maka albarn s1',
  'maki nishikino s1',
  'mikoto misaka s3',
  'naru narusegawa s1',
  'nino nakano s2',
  'otoshimono nymph s2',
  'rena kariu s1',
  'rin tohsaka ubw',
  'sagiri ameno s1',
  'sagiri izumi s1',
  'seishirou tsugumi s1',
  'sena kashiwazaki s2',
  'shakugan no shana s3',
  'shinobu miyake s1',
  'stella vermillion s1',
  'taiga aisaka s1',
  'tsundere',
  'yui kotegawa darkness'

1. Upload this files as a zip to google drive

2. Deploy an RTX 4090 pod

3. Download dataset and pony.safetensor to pod

4. Run sdxl_train.py

5. Download output safetensor

6. Merge output safetensor to autismmix confetti