Note: The showcase uses very little negatives, and no post processing (upscale/i2i) aside from ADetailer!
My first publicly shared doll! Meant for personal use, but liked the results so figured I'd share with everyone.
I've since retrained her with the same dataset but different captions and other settings. The goal was to make her a little more compatible with Photorealistic models. Mission mostly succeeded!
Alienor_A: Works best in 2.5D/Anime, but is recognizable in photorealistic models as well. However, without proper prompting, and depending on the checkpoint, will skew older/asian. Still, as the galleries posted show, she can still work great.
Alienor_R: The retrained version. Different captions, more steps, different learning rate. A lot less susceptible to drift in age/race in photorealistic models, while still performing well in 2.5D/Anime checkpoints. She did, however, lose some curves and about a decade in the process, so for anime, I tend to lean towards the earlier version. You may prefer this one though! More flexible for realism, but look has somewhat changed