It'll also generate just the feet if that's what you want. Give it a simple prompt like "cute feet, close-up, soles". They'll still be in a footjob position though, and some will be hovering over dickless crotches if you don't tell it to add footjob or cock.

It depends on the other lora and the base model, I think. Start with 0.7 weight on this lora and 0.5 on the character lora, then turn up the weight on the character lora until it looks enough like the character. If it deep fries before it looks like the character, turn down this lora as you increase character weight.

Also, with both loras you can add ~1.3X weight to the trigger words to maintain footjob-ness and character-ness even as you reduce the lora weights, which helps the loras coexist. Adding synonyms can help increase concept expression without deep frying, e.g. try it with some of these added: "with her feet on a cock, cock between feet, two-footed footjob, cock, erection, foot focus, toes, pov crotch, male pubic hair". As long as the concepts exist in the base model or the dataset somewhere, they should nudge it just a little each. You can use a similar strategy with traits of your character, even if they weren't trained explicitly in your tagging.

Try turning down or replacing the negative embedding as well, especially for anime characters since it pushes images toward realism, though I thought it still gave the best results on the Cyberpunk Lucy examples. (Embeddings / textual inversion models add together with lora models etc. when it comes to deep frying from too much total add-on model weight.)

It's coming! I took the shoejob data out of this model but it'll either get integrated or get a separate lora depending on whether it really causes problems having it in this model e.g. random shoe soles appearing on feet when prompting stockings.

But also, what kinds of shoes do you need to see in a shoejob model? Looking for dataset I see mostly high heels, generally platforms, some shiny some strappy and some boots. It looks like there are high enough quality sources for rarer stuff like sneakers, flip flops, in between foot and shoe, etc. if The People require it though.

I've got a few images for it that I've left out to try and keep the model more consistent, like you'd start getting a phantom third or fourth foot in there when you're trying to gen just one person. But, I'll see if leaving them in isn't a problem. You can already get close with a prompt like "photo of (two women:1.3) sitting on a couch giving a footjob together, man laying pov, pov crotch, penis, erection, (2girls, ffm threesome:1.3), 1boy" so it might just work.