Changes in V2: Wings are now added (though a little inconsistent), tags have been updated, and I think the eyes tend to come out a little better. Using a weight of 1 now results in better images than what you'd get in v1 if you used 1, but 0.7 still seems more consistent.

Due to the low presence of her high heels in the dataset I'd only prompt them while you're using the aforementioned weight of 1, otherwise they seem to come out as "regular" high heels. Might see about uniquely tagging them in the future. Luckily, for most coomer purposes you can exclude both the wings and the heels, which are the most shaky aspects, and it'll turn out great.

Coming up next: I did the unthinkable and downloaded the game in order to get higher-quality screenshots using the insane 9x upscale screenshot tool within OW2; hopefully this will result in V3 having the best quality yet.

Mercy in her default skin as seen in Overwatch 2! While there's an abundance of her Overwatch 1 design on here (as well as it being promptable on pony), I had a hard time finding her stock Overwatch 2 design, so I decided to give it a shot. Her halo, bodysuit, pelvic curtain, and heels are all able to be prompted.

This initial model does not include her wings, as my main focus was on the bodysuit. I'll try to include them in future versions.

I highly recommend using a lower weight, like 0.7-0.8, and either using the score tags or a style LoRA (or both, I'm not your boss). Obviously, all this in conjunction with upscaling will give the best result.

The weakest point of this model, in my experience, is her eyes. Depending on your prompt, you might have to roll the eye gacha quite a bit to get a desirable result- sorry! It's definitely the highest priority fix for any further iterations.

Thanks for reading, and enjoy!