Over 20GB of training data carefully weighted merges native generation at 1280x1280 and 1600x1600 (Single pixel pubic hair was generated at 2048x2048 close up image to image at 0.6)
Score can be used with standard Character prompting (WD14 and Danbooru).
Natural language has been reinforced BLIP-2 over 100,000 steps.
Neutral realistic recommended for most uses.
-realistic for source_anime or source_cartoon +realistic will limit number of available images, but can be used for up scaling or 2k realistic images.
Natural language can be used with WD14 tags to refine PONY results.
Description
FAQ
Comments (8)
Has problem with eyes
Having 1000's of generations with it I have noticed that on a few images. Certain expression clip triggers like angry or surprised being heavily trained on anime will trigger on what would otherwise be realistic.
@Felldude so...is there a way to fix it?
@kd231002573 -anime eyes, -source_cartoon -source_anime if your going for realistic. I still don't recommend +realistic for most cases and -realistic can be used also even if photo real is the goal.
For cartoons or anime just reverse that
@Felldude thanks!!
Your model is really nice, I rarely have problems with her eyes, thank you for your efforts :)
Thank you.
⠀⠀Let's compare some PONY models 🦄
⠀⠀⠀⠀🌍 See here my detailed test 🌎
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.