Trained on portraits from Elona and intended to create portraits in a similar style.
Summarized Directions: Generate at 512x512 or 768x768, then...
1) For use in actual Elona
Shrink to 122x122, then crop to 80x122 (resolution that game requires for in-game use)
2) For "accuracy"
Shrink to 72x72, then crop to 48x72 (resolution of the original images used for dataset)
Suggested Weight: 0.8
Suggested Resolution: 512x512 or 768x768 (model trained on 768 resolution)
Trigger Word: ElonaFacePortrait
Key Phrases: simple background, closeup, portrait, looking at viewer
Helpful Phrases: [Color] hair, [color] eyes, [long/short] hair
Negatives:
Green background. (It is used as filler for the sample data)
Glasses (For some reason, despite only having like 2 portraits in the training data with glasses, it seems to default to glasses wearing characters when unprompted)