Collection of my Asian female loras characters
Faces are from https://thispersondoesnotexist.com/ some are just a single picture and other sare a blend of a bunch (creating a face_model in reactor).
Description
Source images: x19
Resolution: 512px
Training time: 30min
Learning rate: 2e-5
FAQ
Comments (8)
Could you please post your training config and captions (or how they were captioned)? These are great results.
i used joycaption for captioning , actually from the TripleX script that idk why it is no longer available 😢 but is just basic image captioner, as for the trainer config, you want me to upload .toml?
K3NK yeah, having the .toml would be great!
And thank you, I was able to find a copy of the Triple X script on web archive (https://web.archive.org/web/20250707083027/https://github.com/NSFW-API/TripleX/blob/master/captioners/joycaption2.py). There's also a newer JoyCaption Beta model if you haven't checked it out yet.
fxbn2849211 added the toml files as additional files
I have never tried training on 1 picture, it works well?
is not actually one picture, is one face, then i generate the training images with sdxl/pony and using reactor for replacing the face, then you train on that, 20 images is enough, different angles , hairstyles and so..