Typ
TextualInversion
Veröffentlichungsdatum
2023-03-30
Basismodelle
SD 1.5
Auslösewörter
kyl13m
Kopieren
Versionsvorstellung
Trained on 123 images for 21,000 steps at a learning rate of 0.0015. 20 Vectors per token.
Good hit rate: all images are clearly her, 40-50% are spot-on.
Adding age-steering prompts can make her look a little younger e.g. "(((20y old)))" makes her look about 28, no age prompt produces her at mid-to-late 30s.
Annoyingly I don't think it's captured the mole to the left of her lower lip, although it could be that my -ve prompts are 'airbrushing' it out.
