DO NOT USE THIS ON ANYTHING THAT ISN'T BIGASP V2 ITSELF OR DIRECTLY DERIVED FROM BIGASP V2, IT WILL NOT WORK PROPERLY.

This is a Dora trained on 1000 images (highly diverse mixed SFW / NSFW dataset), hybrid captioned with both JoyCaption natural language descriptions and wd-eva02-large-tagger-v3 booru tags. Generally speaking the goal here is to "stabilize" the overal composition of bigASP V2's raw outputs. Recommended inference strength is around 0.7 to 0.8, but feel free to adjust as you see fit. There is no particular "trigger word", just rather set and forget in terms of how to use this.