due to the lack of kobo images itt, I decided to train kobo hypernetworks
dreambooth model is not ideal due to the size and the embeds still strugle to recreate her fountain hairall pt files can be found here
https://mega.nz/folder/mAJkFCjS#xzCJb5AAI_c-sUe0IhNIFQ1,2,1 layer trained on 31 images dataset with 0.000005 learning rate
the alt version is using [name]. [filewords] prompt template while the regular version trained on kbknr, [filewords]
not sure if that makes any difference at all
comparison images:
https://files.catbox.moe/9z3q0k.pnghttps://files.catbox.moe/2xmk1p.pnghoping other anons with better pc could do further tests to see which one is better
