Not only is it polite to mention the checkpoints used in the merge, but this is also helpful to longtime SDXL users who have years of experience with various NSFW branches. It helps us know how to prompt right away. My first instinct with a checkpoint that doesn't explain it's model genetics is "meh, too much work figuring out what it's strengths and weaknesses are, I'll pass for now".

Works quite well on my end on forge. It doesn't like high denoise (0.4+) or ip adapter with a weight of 0.6+, however, there is a workaround. If you use any other checkpoint and use adetailer with personyolon set to use this checkpoint, it will have no issues.

Edit: forgot pros/cons for this model. Don't have much time using it in text2img, mainly using it as a touchup, or scene change tool in img2img.

Pros: It does skin and people very well, clothing is quite well done, especially shoes which I tend to find very difficult to make on some models. The person seems to look real, without the strange ai look. It goes pretty fast as a stand alone, and depending on the load, fast when using as adetailer. Havent tried using quality tags in promps, so it may not need them.

Cons: it hates ip adapter, especially at 0.6 or higher weight. it also hates high denoise (0.5+ usually) for inpainting, and loves to change composition of the background when used in img2img (adetailer method stops this). No vae is more of a gripe, but still. It doesnt do objects well, such as backgrounds, etc. Using the method below mitigates that though.

Workaround way: load different checkpoint, and load this checkpoint in adetailer.

First checkpoint does the background and shape, adetailer does the person without plastic skin and face. Settings I used for adetailer were 0.3-0.35 denoise (still quite strong, so not really that low for this checkpoint), 1.1 cfg, 14 steps, 100 pixel mask, 8 mask blur, mask set to merge, controlnet set to passthrough with 0.5-0.7 weight, use other checkpoint set to this one, and sometimes clipskip 2. Just make sure a vae is selected at the top. I like using the finetunefailure one, but sdxl vae works too, and if the tokens are above 77, longclipL seems to help.

I must say I am generally very sceptical about accelerated checkpoints. Since I am more interested in quality than speed, I avoid Turbo, Lightning, Hyper, DMD2 and other accelerated checkpoints. This one is an exception, and the only exception so far. The quality, including hands, and overall level of detail are as good or better than many non-accelerated checkpoints. Congrats on a successful merge!