Tena ??????? v0.5 - Realistic ABDL

A LORA for putting pretty ladies in Tena style disposable diapers.

If you like my work you can tip me through Comiss.io.

One day I will figure out how to make one LORA that handles lots of designs, but for now I have focused on just getting one brand looking consistent, with repeatable lilac accents and all. This is not a very user friendly version and it took a lot of work to make the preview images (read below for details) but the results can be very good.

Polite notice: This LORA only depicts adults doing consensual adult things, and fully complies with the TOS at CivitAI.

The trigger is just tena, and I recommend LORA strength 0.8 to 1. It's supposed to respond to wet and dry as well, but it's hard to tell if it does at the moment. Also, this LORA works much better with lower CFG, so knock it down to 5 or 4 (or even less!) and push the LORA strength up if you are struggling. Also prompt for outdoor scenes; it seems to work better with natural light.

The first two preview pictures (in the park) are just right out of text2img. There is some ADetailer involved, but no further work except picking the good ones from a batch of 20ish. I'd say I got 9 good pictures out of that batch. Do be aware though that they were made on CFG5.5. Even at CFG 7 it can burn out, and have artefacts in the background.

The other four (indoors) had quite a bit more going on. The process for making them was to prompt a scene (including the LORA) and then running them at CFG 12 or 14, which made for crappy pictures but solid ??????? and framing, and with a ?????? drawn in the right place. I picked out a decent pose, then fed the picture into ControlNet and regenerated the picture from the same prompt and seed with CFG set to 3 or 3.5, and then tweaked from there.

I got into doing this basically because the LORA messes with the proportions of people's pelvises, and it results in a lot of confusion where limbs go. The high CFG is strict enough to not care, then the controlled version has a controlnet to rely on to tell it where legs go.

I have tried using it through latent couple, which does work, but it is much more labour intensive to make masks as well as controlnets, and you often see it making a good ?????? but that has the shape of panties. You either end up chasing it around with multiple iterations of mask or give up and just start again anyway.