Based on Kroma (teacher model)
All images in the dataset are realistic photographs (including a small number generated by Z-Image Turbo to accommodate various aspect ratios).
Compared to other similar LORAs, this LORA has a smaller impact on areas outside the genital region.
When the prompt includes 'innie vulva', the strength can be reduced to 0.8. (No trigger words are required, but the dataset consists of natural language descriptions that include the above text.)
Description
FAQ
Comments (6)
No example images will be provided. Please try it yourself.
Pass.
Labiaplasty does a pretty good job, as do most of the dozen other such loras. Since your claim indicates you have tested the lora thoroughly in comparison, yet elected to withhold your findings with this message, any hint of curiosity I may have had is quelled by the maxim that an arrogant person is far from an objective judge of their own work.
@firemanbrakeneck Image evaluation is subjective. However, I’m only referring to this particular LORA: when used in conjunction with Kroma 0.31, its impact on the rest of the image is (significantly) less than that of other similar LORAs. This is something that can be objectively determined.
@suede2031691 Ah, thanks for clarifying. I can see why that would be so, given that 3.1 was purportedly trained on the undistilled version of krea, whilst all loras were trained on the published version (as in the zit-zim cross support situation).
I haven't exactly been compelled to try 3.1 after testing the first two versions, but I'll keep that in mind. Though frankly I'd rather expect kroma to be able to handle this concept well enough on its own.
I like it a lot. Thank you ❤️
Everyone who downvoted the OP is a fucking idiot! Why NOT provided? Try using those low-IQ brains of yours to think about it for once!