Style lora for creating anime screencap-like images with emphasis on underwater.

This lora can create decent wetsuits and passable underwater scenes.

It can also try to make scuba gear but that is not so good. Scuba stuff uses tags "scuba", "diving mask", "regulator" and "snorkel". If they bleed in you may use some of these in negative tags.

To get anime screencap style use "screencap" in your prompt. This word was not actually used in training but it seems to work well.

Training details:

The model was trained from 1243 manually selected and cropped screencaps from amanchu!

The images were captioned with wd-vit-large-tagger-v3. The autotagger is not good for scuba stuff so that had to be done mostly manually.