Typ
LORA
Veröffentlichungsdatum
2023-05-08
Basismodelle
Other
Auslösewörter
warspite
Kopieren
Versionsvorstellung
Trained on 149 images with 2 repeats for 7 epochs at a network dimension of 16 and alpha of 8. Seems to me like it's overfitting slightly. Works well enough on AOM3, especially in hires renders.
The model trained too quickly for my liking, I'll try adjusting learning rate and maybe fiddle around with splitting the dataset into a character and costume subset. Experimented with it a bit but couldn't get it to work properly yet. Look forward to that in a future version (maybe).
