Suzanne's XL Mix

Yep, it's got some minor tweaks, and the version here is pruned and fp16 for faster generation. Delete the old one and download this one. I'm hoping for some further improvements sometime in the future with 6.1 with some built-in image enhancements and added detail without using LoRAs.

@Suzanne I wouldn't say I need it, but I'm going to use it to train LoRAs, so I wouldn't be turning any away.

Outfits or particular descriptors certainly have an influence on the style, as only some of the models included know certain keywords or concepts. By default, this checkpoint tends to walk the line with semi-realistic images, but concepts can change that. For example, if your subject is something like "neko cat girl," you are much more likely to go anime, while keywords like "realistic photograph" as well as keywords that describe composition and effects in photography (like describing specific types of lighting, angles, or framing) may pull it into realism. Sometimes it's a weird keyword that seemingly has nothing to do with a style, but for some reason, one of the models was fine-tuned using it, so it changes the style when used. Those are going to be trial and error to figure out.

But here's my general method for creating images. I use four styles (usually a combination of them) to create images. These are the only quality keywords I use, and all of them tend to keep the style neutral to whatever I intend the model to generate. Try to add these to your prompts and see if it helps. I don't use any negatives like "bad hands," etc. These are styles formatted for Automatic1111, so if you aren't familiar, it follows the format: style name, "positive prompt","negative prompt"

Quality-Basic,"high resolution, best quality, 4k","(((watermarks, website, logo, artist name, web page, letterbox, signature, webpage, text at bottom, text in corner, company name, names, labels, captions, websites, comments))), words, text"

Qualty-Portrait,"full body portrait, amazing details, striking eyes, perfect face",

Quality-Dynamic,"detailed background, dynamic angle, dynamic pose, full body shot","centered shot, symetric pose, symetric image, multiple views, multiple panels"

Quality-NSFW,"NSFW, erotic, perfect body, sexy, 20-years-old","SFW, children, loli, child, kids, teens"

I don’t care about the above uses, and could care less what is done with images, as long as uses are legal. I would to be credited for use of the model in the Lora training. That sounds interesting. However, I’ll need to double check permissions on any model I included in a particular version to make sure I didn’t miss an inherited permission that doesn’t allow one of those uses. Are you taking about a specific version?

@IndustrialVectors I don't include any model in any mix that doesn't have more restrictive permissions than what I list on my model pages. Normally, I don't even download them. I double-checked, and no model has restrictions on selling images, and a requirement that merges need to use the same permissions.

Always welcome. Thanks for trying it out!