XenoGASM-MK2: The 2nd Coming (NSFW Semi-Real Portraits and Fetishes)

The original XenoGASM checkpoint has always been my most successful model - and probably for good reason. It was one of the first to start combining concepts previously only seen in Comic/Anime/Hentai models with semi-realistic looks and feel. The big (self created) problem with it as updates came was that I was focusing more on "making it look more real" than focusing on "making it do more unreal things in a semi-realistic way."

And so, with this new understanding of what I did wrong with that, combined with many things I've learned about merging since that first release came out some 8 months ago, I present you with: XenoGASM-MK2: The Second Coming.

Why A Separate Release Rather than an Update? I ended up NOT using V5 (or anything after v2.1 actually) in this. It's basically a cleaned up and branched fork from that original. And since the goal isn't "total realism" but "realistic unreal things" it's a new direction - hence new model.

If you would like to read more about the whole story behind this madness, please check out my Behind the Scenes and Tips for Xenos 9th Generation Checkpoints article. After the boring history bit, there are some good tips and list of some of the unique and fun things this (and my other Gen9 checkpoints) can and like to do.

What is XenoGASM-MK2 In a Nutshell?

• Semi-Realistic (and almost photo realistic with some general prompts) but with a vast knowledge of styles, concepts, characters, and more. Many of the things this checkpoint can do (with the right prompting, anyway) are things typically reserved only for Anime/Hentai/Toon type models.

• Style Pulling rather than Switching: When you call a style it does NOT switch to that style, but instead, it works in the opposite way. It won't make a picture in the "style of vincent van gogh" but it will take style elements, colors, and other aspects of that style and bring them in to the "semi-realistic" style. (Examples of this are in the Behind the Scenes Article in This Section.) This should hopefully make future updates easier and more fun, too.

• Awesome At Fetishes and Sexy Stuff, Good at Sex Stuff: Fetishy stuff (especially the weird shit I'm into) works really well - mind control/hypno, transformation themed things, and so on. It does an A- job with skin and sexy parts (though better than many checkpoints, it's one area I really want to improve on - nipples. Ugh... they're annoying. lol) And while no checkpoints are "really" great at sex positions (beyond maybe a few basic ones), this has been memory optimized and "primed" with a set of base concepts that feed nicely into the various sex LoRas and sexual concept models out there. And, because it's always wanting that semi-real look - most LoRas aren't going to cause as big of a "style shift" as they would on many checkpoints.

• It's just a Damned Good Model: My hit/miss ratios were almost 100% hits on the basic person, clothes, setting type prompts. Now, I won't say that there aren't occasional hand issues and whatnot, but... they don't come as often and you can get more dynamic with your posing and whatnot before weird things like tentacle fingers and all that starts to happen. And boy oh boy - just plop a name of someone into the prompt and see what comes out. It's just good, it's just fun. Simple prompts rule, but carefully crafted complex prompts can also make some amazing shit.

Some Considerations As You Start To Use This

Ahh. Here comes the fine print after the list of features. Honestly, though - some of these are not "downsides" so much as just having to do or approach things a bit differently than you might with other models.

• Open Connections: In order to make things work which normally wouldn't work together... like say "spoken hearts" (or speech bubbles at all) in a somewhat realistic generation instead of a toony one... I broke down a lot of connections between things which cause bias. So, while it is still aware of the fact that you can't "really" see love hearts or see what words people are saying - it doesn't say, "Well, we can't do that speech bubble because this is a realistic photo" as strongly. You can essentially coax that out.
  
  The downside of breaking those connections is that it ends up losing some of it's other sanity factors. So, for example, on most of my other models, if you just say "a woman walking down the street" - they will almost certainly put them into appropriate clothes for walking down a street. Those rules being greatly lessened also means that this model is just as likely to make them nude or dressed in something inappropriate for the setting as it is to get something appropriate. So you may find you need to prompt more for specific clothing then you're used to.
  
  Many other concepts and things work similarly. Don't assume it will do something smart like the other models. But you CAN probably assume it'll do something fun and interesting!

• Reverse Connections: A lot of things in this model work backwards from what you're probably used to (unless you do a lot of stuff on some of the recent photo real models that use similar methods, or are at least achieving similar results in some way- I'm not actually sure how they approached it). As mentioned above... when you call a style, it pulls the style to you rather than switching you (and the whole image) to the named style.
  
  You CAN of course, up the weight of those style calls (e.g. (Van Gogh Style:1.4)) and put a punch of "realism" tokens in your negative prompt and force a shift - but there are other models meant for changing styles so you'll have an easier time with those. The "bring the style to your semi-realistic stuff" is a (somewhat) unique/new feature, not a bug. :)

• Lose Your Default Prompt Tokens: This Works Differently. This model doesn't really think about everything in the same way as other models. Styles are more about shapes, colors, the types of things and looks included than the medium in which the style was drawn/created. And with logic connections breaking down - many negative embeds, positive embeds, and other "quality enhancing" things might actually be working hard against you on this model.
  
  Start with "no default" positive or negative prompts and build it out fresh. Then use your "bad hands" negative embed if, and only if, you find yourself in the situation where hands are going bad.

• Man Hand Syndrome: If I have a complaint about this (and my other Gen9 models starting with XenoGRAPH) is that they are MUCH better about not having spindly hands and other weirdness. (This is creeping up in all sorts of new models now - so this tip might work for other things, too.)
  
  Those weird spider hands can still come, but you have a lot more you can be doing in the prompt before that happens (usually). But what DOES happen before "really bad hands" creep in is that the hands look big and manly rather than feminine. So if you are working a prompt and start seeing man hands... see the fix:
  
  Fix: "perfect hands, feminine hands" in the prompt - sometimes just one or the other are needed. You can add some weight to them sometimes to help prolong the amount of messing around you can do before breakage, but understand that as the weights go up on those tokens, so do the model's tendency to focus on the hands. So it can be a balancing act.

Learn More/Do More

Please don't forget to Follow Me here on Civitai so you will be informed of new things. (Plus, as of this writing, I'm just under 250 followers short of being eligible for the next round of Buzz Cash applicants - so every follow counts!)

Need help prompting, finding tools and techniques to accomplish specific things, or generally want to learn more about SD, Generation, Making your own LoRas, Checkpoints, TI's and more - join the Xeno Engine Discord - we're a growing group of creators from all over the world with various, though often a bit deviant, interests. (NSFW).

Please take a moment to like, collect, and review any of my models that you enjoy. Each of those things helps the model be more visible to the community. The more visible, the more likes and reviews and downloads, which leads to more buzz and more rankings. Eventually, I hope that my earning pennies for each dollar put into buzz will get me to the point where I'm making an extra $10 a week to cover my Friday PBR Six-Pack. I know... high hopes, but... I can dream, right?

The Only Dumb Question is the Unasked Question. This is an important rule that I live by. You might feel shy about asking for help with something that is probably obvious to most people who have been doing this for a while, but here's a secret... we ALL were at that "total newb" stage at one point - and heck, when it comes to prompting especially, I still consider myself a "somewhat skilled novice". If you want some help on something - it's only dumb if you don't ask. If you do ask - on site, on discord or wherever - you'll never be looked down upon by me (or any of the people on the discord, lest they be shut down like an illegal gambling operation.