All models are based on epsilon prediction version of noob v1.1.Version 1 - the only thing it does is improve foreskin detail. It does this very well, but doesn't add any other featuresVersion 2 - adds "foreskin suck", "tongue inserted in foreskin", "finger inserted in foreskin", "foreskin grab".Messup versions - version 2 I tried to add objects in foreskin but it is hard to prompt things that weren't in the dataset. Also I think I messed up the weights somehow and it may mess up other tags.Version 2 is an extract of the following finetune/checkpoint: <a target="_blank" rel="ugc" href="https://civarchive.com/models/1960360">https://civarchive.com/models/1960360</a>I have gotten the best results with res_multistep_ancestral + simple. I'm not an expert on samplers so feel free to experiment.Version 2 sample prompts:<ul><li>1boy, nude, foreskin, sitting - get a basic nude male with foreskin</li><li>2boys, foreskin suck, fellatio, pov, pubes, solo focus - get a point of view of a male sucking a foreskin</li><li>2boys, foreskin suck, fellatio, pubes, full body (+solo focus, pov in negative) - more likely to get a full body image</li><li>2boys, tongue inserted in foreskin, foreskin grab, pov, pubes, solo focus</li><li>1boy, nude, finger inserted in foreskin, sitting</li><li>1boy, nude, foreskin grab, sitting</li></ul>It is not perfect. I need a larger dataset to make this more consistent.

NoobXL Foreskin

version 1 adds foreskin, but is not consistently able to cum from the opening.

Files with SHA256: 3b4b7d82510e40b85cc7d3e0de7f5edb5fd0375b7db3320daf6a1e5bdc1bcf54