Includes moves 1 (180 images), 2 (118 images), 3 (96 images), and 3+1 (213 images).
Almost all characters were tagged. But only the ones with lots of images work reliably
Characters that work well. You can get their likeness without any effort: souryuu asuka langley, ayanami rei, makinami mari illustrious, ikari shinji , katsuragi misato.
Characters that work somewhat. You need to describe their appearance in detail to get them: akagi ritsuko, ibuki maya, kitakami midori, horaki hikari, nagisa kaworu.
Useful keywords:
Each movie had separate tag in captions. You can use "eva_1", "eva_2", "eva_3" and "eva3_1" to slightly nudge the style/character to better align with specific movie.
Including "(anime screencap:0.7)" helps in avoiding situation where you get face in style of the movies but body in a general style.
All images were tagged as "evangelion" but that doesn't seem to be helpful keyword to use.
The model managed to learn how to do interior of the EVA cockpit, which is nice.
Training details:
600 screenshots. Trained for 10 epochs on illustriousXL_v01.
