Sweet AI
patreon
SWEET LyCORIS | Elizabeth Katz | V2
🕑 Added 2023-08-01 23:17:30 +0000 UTCEthnicity: American-Jewish
Update to what would i consider the perfect girl, trained on 80 images (four different women), since i'm a GUI guy i'm stuck with NMKD for training which is outdated as fuck (he removed inpainting on the new version, which worked absolutely fine on 1.9.1 even on 'non-inpainting' models), it does not even include epochs, tried many aproaches but it nevers get close to what people can produce using WebUI trainers, but oh well what ya gonna do, have to wait for someone to step in and make a competitor for NMKD
Trying to learn a bit of LoRAs and LyCORIS, but every single tutorial is either for 1111 or koyha_ss and that sucks massive cock (if i get a error first try using any cmd then i bail), the only thing that seens to have effect on trainings using NMKD is the learning rate everything else don't seen to influence on training, also other people's LoRAs are 100mb+ i'm not sure but i also believe this influence quality, not sure why NMKD is stuck at 60mb, but i'm a visual person, nodes and the fucking WebUI design is so....ugly to me that i will be stubborn and stay with NMKD
As for my girl, well i'm close to the look i want (well she is lol), removed the faces that were influencing to much on her jaw and nose, but her nose is not quite the way i want, i want it to be larger and her cheeks to be more fat/chubby and add freckles on her cheeks, but there is what, 10 women that look like what i want in the world? i honestly wish in the future AI trainers would allow us to make selective choices, like mix and match, jaw, mouth, nose and eyes.
This LyCORIS is NOT tested using txt2Image instead it is focused on img2img using any WebUI or GUI that supports strength influence
Tested on CyberRealistic - v3.2
Recommended strength: 5.5 -6.5
Recommended weight: 0.8-1.0
Trigger Word: SweetAI_ElizabethKatzV2
Tested on 1408x1408
