If you don't see any images, it's because the automoderator is freaking out, probably because the prompts contain 1girl.

• This is a DORA file. It may not be compatible with your inference software.

• This DORA was trained directly on GurilaMash-v2. It is designed to work best with that model. It will likely not work well on other checkpoints.

• Your Buzz Donations Will Go Towards FLUX LORA Training :)

Reverse Cowgirl Position

for GurilaMash V2

Alright, so...first, yes, generally speaking, this lora (dora...whatever) shouldn't even be necessary. Most porn models do a good job with this on their own. So, where's the value in this?

Well, In my opinion, this dora is capable, with detailers, and maybe a touch of upscaling, of producing the most photorealistic images of sexual activity possible with SDXL. Faces, facial expressions, genitalia details, skin details, the whole enchilada. I challenge anyone to do better.

Now, this thing isn't perfect. To squeeze out the details I wanted, it's a few steps overtrained, and the consequence of this is hands. 2 out of 3 generations will have bad hands. That 1 out of every three, though? Should be money. Try it about .5-.7 for best balance of details without too much distortion. Maybe even lower, revcow is heavily favored in most model datasets.

I've uploaded 2 separate models. They were trained at the same time, with the same parameters, using different datasets. I have A LOT of reverse cowgirl images, and the quality was better using a smaller dataset, but, i didn't want it all to go to waste. There are subtle differences, and Coar2 has a touch more variety, including more 3somes in the dataset. Don't use both at the same time, but, play with them. You'll see what they can do.

Ok, now for a little musing if you'll humor me. This is likely my last SDXL model. I have many more datasets to train. But the FLUX revolution is happening faster than is even believable, and I'm not certain it's worth the time and expense to perform further training on SDXL, which, honestly, doesn't make me too happy. I have hundreds of hours invested in this. But, as we can all see, FLUX is going to be badass for NSFW, so, that has me pretty excited. I'm probably going to wait until the FLUX NSFW models get actually good, and start working from there. I've got over a million images, all captioned and organized, ready to go to the trainer, so lets see what happens.

If there's something specific, and not too fucking weird, you can ask, and maybe I can get motivated to cook a lora or dora if it's something I have a dataset for. Never hurts to ask. As it is, I'll probably burn the Buzz I've got on a few datasets. If anyone thinks there's a concept that should be a priority for FLUX training, let me know in the comments and lets see what we can do there.

Thanks for all the support. We're not done, just retooling.

as always:

score_7_up, photo \(medium\), realistic, photorealistic, reverse cowgirl position,

That will get you where you want to go.