Krea 2: I spent some time on this one and I think it's one of my most capable concept/style LoRA yet. It focuses on anatomy, material, lighting and NSFW interactions - it's hard to say if it's more of a concept or more of a style but since it can generate object and action specificity I'm labeling it a concept. It can be used to create NSFW images outside a bathhouse as well as completely SFW images if you want to use it for stylistic purposes. It also works well with stacked character LoRAs (some examples in the user submitted gallery).

It was created an entirely upgraded dataset to not only generate very high quality image outputs, but also get that moody, saturated volumetric lighting with deep shadows with lighting that favors deep reds, blues, purples, amber, indigo and the like typically seen in such unseemly venues. It's not stuck on only being able on that color grading, but will lean into that, so adjust your prompts with lighting specificity if your looking to create a different type of atmosphere. The image gallery has example prompts to help if you need it. Bypass LoRAs and/or Comfy nodes shouldn't be needed - although as always with Krea 2 they do help.

It doesn't need other NSFW LoRA helpers or checkpoints, it can pretty much generate whatever debauchery you can think of that would go on in a bathhouse on its own. I trained it on nearly 1,000 HQ images, so when I say anything you can think of with this concept, I mean it. Why so many images when Krea 2 easily learns a concept or style? The reasoning is that I wanted this LoRA to also work well with stacked character LoRAs without altering them and give maximum versatility in general. If I went with a lower number, I'd need to crop faces out of the dataset and the generated outputs would tend to create face cropped images, or risk this LoRA messing with character identities.

I also incorporated some of stuff of my other published NSFW concept works in here, plus a few a few extra just for grins.

This one took some time to get right, with treating the dataset and new captioning strategy and a few trial training runs so I'm making it early access to potentially recoup some of that time and effort. You should be able to test out some test generations before you decide to pay. Hopefully I set a price that balances the effort and product.

Pony & SDXL versions:

.. Have you ever wondered what's up with that building with covered or no windows, a long line of just men on weekend nights or special event holidays, looks like on the inside? Well wonder no more my friend! lol

This is a different type of LoRA I haven't trained before - background / building concept. My SDXL version seems to be fairly stable and produces consistent results.. still tweaking the Pony version.

You should be able to specify color lighting schemes, various interior design elements commonly found in them, etc. It has mixed results with stacked LoRAs with the Pony version more stable in that regard.

Model based on several different 'men-only' saunas in US and Europe.. allegedly, since I've never been on account of my strong moral values and chastity!

main trigger word: gaybathhouse