"Every picture tells a story, doughnut?" - Rod Stewart
Almost a month on the making, this experiment began as a LoRa Training experiment that went horribly wrong. It then brought me down a road of looking at old checkpoints, finding out which sloppy LoRas that had been merged in (because, at the time, that was the best option) and negate them, clearing room to let the newer/better training from the newer models work without that older stuff getting in the way. (If you want to see the whole crazy process... check the dev blogs on my discord (link in the Learn More/Do More section below).
The Result
This is a general purpose checkpoint. Most of my models, to date, are portrait oriented and I have focused a lot of energy in the past on making great looking people. Backgrounds were always pretty good - but when you start to combine portraits with complicated scenes, the scene just looks flat or unfinished. This balances the two.
Less energy was spent on "optimization for memory and speed" and more on "getting the most out of every prompt." As such, you might need more steps than my other checkpoints to ensure the job gets finished. I did most of the preview images with LCM in my AddNets. For simple prompts, the 8-12 steps works fine, but when you get more complex or if you're using the XD-Scene Embedding (see below) you might want more. I was getting great results (from low resolution 512x starting gens, then upscaled 1.5 times) with 18-20 steps. Larger sizes and different samplers were similar. So - use your own personal favorite settings for generation, and if things are looking a bit unfinished - add some more steps and see what it does.
Styles
The default look of this (most of the time) is a sort of "realistic like a modern game engine" look. As you shift to fantasy/sci-fi themes, that style tends to push more toward fine art/fantasy painting look. It drifts quite a bit based upon subject - so declaring a specific style in your prompt rather than just describing the subjects will help with your consistency. Style LoRas seem to work really well, too.
This model knows a lot about games, movies, television, and pop culture type things.
It doesn't quite go to photorealism, but when you start talking about photography or cinematic things - you can get it pretty darned close.
Beyond that - you'll need to just try and see - this really is a general purpose model and, while I felt like I wanted to tweak and play with the prompts to improve some areas, most worked pretty well right out of the box. Of course, complicated concepts are still complicated concepts.
You sometimes need to add some extra wait to artistic styles to get them to "break" the tendency toward semi-realistic 3D game engine stuff. Fantasy and sci-fi themes make it easier, but just going from "(vector art:1.1)" to 1.3 or 1.4 weight (and maybe combining some other art tags in there) will usually get you rolling. At lower weights, you'll see some of the style "elements" kicking in, but the style medium itself can take a bit more punch.
This will trend toward SFW output without any issues most of the time. That said, if your subject is a famous Victoria's Secret model and you don't describe the clothes, they'll likely end up in sexy lingerie based upon that subject's training over the model's bias. Talk about sexy things - it'll go sexy. Don't and it'll TEND to stay SFW for you.
I'll leave the rest of the playing up to you. Treat this model like any other general purpose checkpoint and let me know what you think!
Companion Embeddings:
You can download a default negative embedding and a cool positive "scene building" embedding here: https://civitai.com/models/381513
The NoDrama-neg embedding does what it suggests - and just blocks, discourages, or redirects certain bad tendencies to encourage the good tendencies to stand out. So far, I haven't found that it prevents any of the things I've been TRYING to make the model do - but there's a LOT of stuff in there, so if you're playing with a new concept or style - I'd recommend leaving this out until you dial in the prompt first, then bring this in near the end.
The XD-Scene embedding is sort of cool. It's basically a bunch of "thematic <scene, clothes, background, etc.>" type tokens along with some to enhance "drama". It's designed to let you start with a simple prompt - and then this will flesh out the scene. In some ways it's similar to an add_detail type embedding, but this adds "things and moods" more than "details to what's there".
Note on Preview Images
While I did a TON of early testing with different resolutions, samplers, and so on, most of the preview images are lower resolution "proofs of concept" and NOT "finished works". I've tried to create things with a variety of subjects, themes, and so on - and then just posting whatever comes out - by looking at the previews, you might get an idea of some of the things the model struggles with just as much as what it does well.
I make the checkpoints and try to give you tips on fun places to explore - but it's up to YOU to make the great art with it. I'm just the tech nerd.
Learn More/Do More
Please don't forget to Follow Me here on Civitai so you will be informed of new things.
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.