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".Version 2 is an extract of the following finetune/checkpoint: https://civitai.com/models/1960360I 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:1boy, nude, foreskin, sitting - get a basic nude male with foreskin2boys, foreskin suck, fellatio, pov, pubes, solo focus - get a point of view of a male sucking a foreskin2boys, foreskin suck, fellatio, pubes, full body (+solo focus, pov in negative) - more likely to get a full body image2boys, tongue inserted in foreskin, foreskin grab, pov, pubes, solo focus1boy, nude, finger inserted in foreskin, sitting1boy, nude, foreskin grab, sittingIt is not perfect. I need a larger dataset to make this more consistent.
