I don't know how you create this, but this is fucking amazing. Good job!

could you make a sd1.5 version？

Any clues how to make images more precisely? Most of the time I have got the results when everyone just standing still without any sexual action at all. What can have an impact on that? I use same prompts as images above. Only when I copy seed from them I get what I need, but whitout seed but same promts, 90% of the time generates just standing girls with no sex.

I haven't successfully achieved a single nonparticipant. I've looked at the training data and there is so little variety in the tags that you would actually need to achieve these results. It makes no sense that the training data is not properly tagged to reference what this advertised it would do.

e.g. ZERO images tagged voyeur, or anything else in the realm of what is advertised: zero tags for caught, walk in, waiting, observe, watch, etc.

The actual full list of dataset tags, in descending order: multiple girls, hetero, sex, breasts, 1boy, 2girls, nipples, sex from behind, nude, long hair, blush, blonde hair, doggystyle, penis, brown hair, 3girls, black hair, dark skin, large breasts, thighhighs, school uniform, short hair, skirt, dark-skinned male.

How does any of that help achieve the result? It doesn't. Even when I try multiple tags from that set I still get nothing close to a composition that makes space for anyone to appear in the background.

This page needs better documentation on what models are supported, because as-is the LoRA introduces no new words and gets completely ignored in favor of what the prompts already meant in general.