Creates the likeness of the actress, Sydney Sweeney
Guidance of 4-6 works best.
v1.2 is trained at 512px with the keyword sw33n3y, it is trained with better data at a lower learning rate, and it significantly improves issues with glitching and frozen faces.
v1.1 is trained at 1024px with a keyword (sw33ny) (yes I spelled it incorrectly, it was an accident)
v1.0 is trained at 512.
Description
Trained at 1024, and with a keyword: sw33ny
FAQ
Comments (8)
Great Work As always
new version is way better, way way way way better
Thanks so much for this. Are you captioning the images, or only using a single text caption "sw33ny"?
so is this usable with some other lora...lets say for example something more action based
tested it with some of dtwr434's "action based" loras and it definitely works well, usually keeping the character lora at 0.9 weight and the action lora at 0.6-0.7
@vslinx indeed, cranking the CFG up to 10-11 helps remove the banding as well as help compensate for the lora strengths being low