Trained on high resolution sharp images of amateur female women (real photographs not ai) opening mouth wide, sticking out tongue, showing uvula, displaying teeth, also a few nudes to get complete body coverage.
Using this lora will improve the texture and display of teeth, tongue, lips and (facial) skin in general.
No trigger words, just use your prompts as usual and drop this lora inbetween your usual workflow and enjoy.
Lora strength between 0.8 and 1.5, the lower the more the original image is preserved but also improvement is less visible, the higher, the more the facial image leans towards the training set (caucasian white females, open mouths)
Tip for enabling/activating good NSFW: Install https://github.com/huwhitememes/comfyui-krea2-conditioning (the improved one) then use node Krea 2 Conditioning Control with preset: balanced, per layer: 0.8, 0.8, 0.9, 0.9, 1.0, 1.0, 1.2, 3.0, 6.0, 1.5, 5.0, 1.2 , multiplier: 5 and renormalize: true. The samples in this gallery are using this.
Works well with other (character) loras.
Description
FAQ
Comments (7)
Krea seems really good at doing this without LoRA
i would add also that for realistic braces it shouldnt be on perfect teeth.
yes of all the models so far i think krea2 is the best equipped already. I do have some issues with textures sometimes (really pixelated), so in that case throwing in this (or other realistic) lora can never hurt to try, it's free for you (for me it took some invest ;-)
@MetaGen to be fair you are right but my training set really did not have really misformed teeth even with the braces, would be good to try to find and add for next time, nice tip
adds some extra exaggerated mouth effects when compared to base which is nice. It exacerbates the original krea2 base problem of dry and overcooked textures on closeups which is what we need fixed with a lora. Keep up the good work!
indeed that's the issue i noted the 'bubbly' textures on closeups too often for my taste on the plain base/turbo; first i was not sure if was worth cooking this lora, but i am sure this lora can help in that esp. closeups and high-res. just play with the strength as indeed it was trained on solely open mouths and so tends to inject that pose on higher strenghts, even if not explicitly prompted.
Hmm, to me, tongue and braces look almost more realistic and convincing without this LoRA. But something about my setup doesn't seem quite right anyway, because the images suddenly look extremely trippy and psychedelic — which I actually think is pretty cool. Unfortunately, it distorts text in the image a little. Anyway, thanks so much for sharing!