Which LoRA have been merged into your model? I ask because, so far, none of the NSFW KR2 models are consistent regarding vaginal penetration.

The official KR2 model was censored, and after the censorship was somewhat lifted, a major flaw resulting of its training appeared: it forces anal penetration at the expense of vaginal penetration.

Regardless of the model, i seek consistency and prompt adherence. It musn't force stuff i haven't prompted. Therefore, due to this major flaw, i am forced to use a dedicated LoRA, SNOFS, for everything related to vaginal sex. Your opinion on this matter would be appreciated.

Thank you for the model - however the int8 quant seems to have the known issue with missing metadata which at least ForgeNeo requires for MixedPrecision. For example the early Cyberrealistic models by Cyberdelia had the same issue - in the meantime he created a node to create int8 quants with the correct metadata format: https://github.com/cyberdeliaAI/ComfyUI-Cyberdelia-Quantizer

He (Cyberdelia) can also provide a script via PM to fix an existing int8 quant without having to use the Comfyui node - this was successfully used by @daceheg192491 to create a corrected version of his dmergev3 model.

Would you consider creating a fixed int8 version? Thanks!