American pinup, glamour and fetish model.
1.5 version: For best resemblance use together with the pair embedding (LINK) and adjust weight of the LoRA down if needed.
SDXL version: Trained on a different dataset.
Flux version: Trained on yet another dataset, works with NF4 version of Dev.
Description
FAQ
Comments (5)
any plans for making an SDXL version of her?
I don't think I can get enough HQ images for XL training - images from this era were quite small, resolution-wise. But I can look at it, it's not a bad idea.
@dolirama126 does SDXL fail so badly if you feed it loq resolution images? what if one upscales them first?
@barlogspam176 Frankly, I don't have too much experience with XL training (as my GPU is only 6GB) to answer these questions. Upscaling is likely possible, but it still adds unnecessary artifacts. Training with lower res images might work but AFAIK it's not really recommended...
@barlogspam176 Well I tried it with low res files and onsite trainer, it's not too bad although I still prefer the 1.5 version. But XL can be useful for some of you, hopefully :)