Update V2 Portraitfocus: as the title says, minimal tagging also. normal v2 should be more flexible.

Update V2: Expanded Dataset and added cogvlm captioning. Still some occasional bad seeds, but works pretty well generally.