This model appears to be capable of just about anything you ask of it. SFW, NSFW etc. etc. - It also follows the prompt rather well compared to many models. It looks fairly realistic, perhaps not the most realistic, but it doesn't suffer from the close-up syndrome that seems to plague many other realism models.

At 1024x1024, and similar resolutions at different aspect ratios, it produces good results. However best results are seen by further using it to refine the upscaled image. Somewhat like the hi-res Fix feature of Automatic1111. Almost all defects that may occur in anatomy (eyes/fingers) are corrected simply through this additional step.

I've not really tested it outside of realism, but I assume it is more than capable of any other styles also, much like most SDXL models with the right prompting. Considering how well it follows the prompt anyway, I doubt it will be much of an issue for anyone.


Support me here!

https://www.buymeacoffee.com/tripleheadg


--

This is a custom merge, the details of which I cannot remember fully. However, what I do remember is that I started with XL13 Asmodeus from Afroman and merged it with dreamshaper using the "add difference" option alongside a 3rd model. I can't for the life of me remember which model specifically it was - either protovision or maybe it was SpectrumBlend.

Either way, afterwards I took that merge and then merged that with Randommax's dark offset lora.


My intentions with this are to use it as a base model to train on top of. But the results it produces are so decent that I thought it would be a waste not to provide the community with it.