Hey you all and welcome to the ultimate journey in SDXL realism!
IMPORTANT NOTE:
The models that will follow on this page are NOT plain merges of some popular or already overtrained models, but a result of a year-long fine tuning and testing of an original model that I use daily for professional purposes (along with other tools, of course).
This will be the story about evolution, so most of the models that I'll upload will be the next step forward (according to my needs at that point in time), usually on top of the previous one. There won't be any spamming or making of "daily" models with a purpose to earn cheap clicks.
The full story behind the veil and all improvements will be posted along each model in the "About this version" section on the right side of this page >>>
Despite some different model names, which represent slightly different approaches, they will all be posted on this page, under the Odd Reality Evolution moniker.
Please support my work if you can: 0x4b320521e2d3ff7eb572ca0d10b29d0b2382d6a0 (USDT-ERC20 network)
Why crypto, you might ask? Because the usual donation and support sites (and all the subjects in between) are taking WAY too much money for themselves!
Thanks!
Let's do some work now. Since this is the first public model after many previous internal ones, I'm posting the description I wrote for myself when it was made... some things may sound fuzzy, but the first two models that I've uploaded were created in February/March 2025, so I don't remember much more info besides what I wrote at that moment.
Here are my "what's new" writings about the first model made many months ago:
Focuses more on human characters than artistic/fantastic details
Better subject/color separation (less material/color bleeding)
Better anatomy overall, with more natural facial expressions
The best possible hands compared to other SDXL models (even better with detailers + proper LoRA)
Usually needs a bit more steps than with similar Sampler/Scheduler combos used before
Sharper, more detailed and balanced
Better prompt adherence, both positive and negative
More realistic lighting and shadows
Better background stability with more details
Much more NSFW than previous models (including new elements and poses), frequently requires clothing descriptions to cover the body (note: this "flaw" was corrected later, among other stuff)
Samplers/Schedulers: [DPM++] 2Sa AlignYourSteps/Karras @35-50 steps - 2M/3M SDE Karras/Expo/Poly @40-50 - 2Sa KL Optimal @40-45
Medium CFG values should be your starting point (6-8), and you can expand and experiment from there.
Hands, Face, Body parts... detailers are VERY welcome, as they can significantly improve the final images! (note: as far as I'm concerned, detailers are a MUST with any SDXL model)
Sample images may not represent full potential of the presented model, but are just some good examples from my testing database.
Please have in mind that the latest uploaded model is not the best one in this line, but just one of the models I've marked as valuable "evolutionary step" in my long journey. Many more will follow...
Since I've put A LOT of time in evolving these models, PLEASE don't sell, merge, share on other sites, or make any kind od mess with them :)
If you have anything significant to say or ask about my models, you can DM me here, and we can continue later on Discord if we find the topic interesting for further discussion.
Cheers and see you again soon...
Description
FAQ
Comments (10)
really really great i must say，even better than those high score checkpoints.
keep going，support u ，really thx :D
Thanks!
I made these models for myself, not for clicks or "achievements", so maybe that's the reason why they are good... ;)
Just try to follow the recommended settings and you'll get great images.
Really awesome checkpoint. Better than all of the "most downloaded".
Thanks!
Unfortunately only a few people here see the things as you do. The vast majority of users are downloading only the hyped models, so I'm thinking of not uploading anything else TBH... what's the point if nobody cares? 😊
@OdderPhase Haven't tried the others yet, but Exoteric Nightmares v2 is really an outstanding checkpoint. I usually don't get great results from pure SDXL models, but this really doesn't feel any different than using a really good realistic Pony or Illustrious model. Prompt and LoRA adherence is hands-down one of the best I've ever seen, and it's quite a capable model on its own, producing excellent results with little to no fiddling with settings. Please reconsider not releasing any future models. I'll certainly be using this one again in the future and would really enjoy trying out any newer revisions you come up with. Great work!
@AFD_0 Thanks AFD, the main reason is pretty much low traction (first model had 700+ DLs, second 500...so it's falling, instead of rising). The other reason is actually CivitAI, which is working like crap, getting stuck (not updating the model pages), ppl have all sorts of problems and that's also why many go for overhyped models only (oh yes, their pages are working properly...guess why 😉). I literally don't get anything for uploading my models, and they are making A LOT of money (by also promoting their "friends" in all kinds of weird ways).
Maybe this can even get me banned, but who cares... 🤷♂️
@OdderPhase You're not the only creator that has recently told me they've been dismayed seeing their newer models getting very low download numbers. Thought it was just everyone catching up after the long holiday break, but think we should be far past that. With other recent CivitAI changes, users just aren't sharing their results anymore either (myself included other than a few personal requests).
I honestly have no idea about the monetary aspect of everything around here, but even if you don't earn a penny from this checkpoint or get a ton of interaction, imo you absolutely did the right thing by sharing this model, simply because it's that damn good and it would be a shame if others didn't get to enjoy it. And no offense to any other popular creators, but I think @hujakuja is 100% right and this merge is significantly better than the vast majority of other realistic models released in the past few years. No idea what's all in it, but I'd easily rank it as one of the top few SDXL-based models (including Pony/Illu), which is no small feat.
I do understand the disappointment, but hope you realize that it's better to share something simply because it's incredibly good and deserves to be shared.
@AFD_0 Regarding all those outages and "dead" model pages on Civit... I've posted a ticket to their support and you know what happened? They told me I'm not the only one affected and they know the problem exists. No solution, no explanation... and then they closed my ticket without me being able to say anything else. Some 40 mins later, my page was miraculously "unlocked", but yesterday it became locked again. Only (some) notifications are still going through. BTW a few of my followers told me they haven't received any notification about my new model.
When I posted my latest model, I have closely followed some of the hyped SDXL models (yes, those that have "new updates" every 2-3 days...the obviuos clickbait scam). ALL of them were working flawlessly, even one that was by "pure coincidence" uplaoded like 5 mins after I did my upload. My page was dead with everything stuck at zero for 2 days, while their page worked like a charm, collecting downloads and showing everything properly. Does that look to you like a fluke? Naaah, they know very well what they are doing > deliberately downplaying everyone, but their loyal money gatherers or big spenders. I'm 100% sure all will be fine with my account if I paid them premium membership or have a deal with them for my models ;)
@OdderPhase Another thing I've noticed lately that could be negatively affecting visibility, is just the sheer amount of crap constantly getting bumped to the top of the line. To explain, every other day, I leave my filters sorted to newest, LoRA + checkpoints, SDXL + Pony + Illu, just so I can see what's new for what I'm interested in. Now everyone and their mother is making ZiT and WAN versions, so every revision of their model (or LoRA) gets flagged as "updated" and I'm forced to see posts for content that shouldn't be shown again to me. It quite literally takes me a good hour or two just to quickly browse through the "new" content because I'm seeing content that really isn't new to me, just "updated". The fact that I even saw your model floating in a sea of crap is in itself quite unlikely now that I think about it and likely due to the fact that your sample image was quite nice (find myself trying to scan everything very quickly, ignoring the insane amount of anime stuff, looking for realistic stuff).
And just an FYI, but if your model has a decent amount of Pony or Illu involved (which I think it might), I think you might get more visibility tagging it as such versus SDXL (results "look" like SDXL, but working with it "feels" like Illu imo). I regularly follow an account that does samples for only Pony and Illu checkpoints, and often find myself seeing models that I missed at first glance.
@AFD_0 I get what you're talking about, but I honestly don't have the time or nerves to play their game, and get absolutely nothing for "winning" occasionally 😊
I made all of these models for myself and TBH wanted to see will they get any significant traction or appreciation in general. My "quality standards" are (un)fortunately way above most of what's happening here, so I now find fiddling with Civit pointless. I don't really care for clicks or shitty buzz or anything that's counted as "currency" here, but seeing my own page being stuck at zero for days (or whatever number when it stops updating) is a nice warning to stop messing with it. It turns into time wasting adventure, which is really appreciated by 10-ish ppl in total. If you wanna talk, ping me with DM, cuz we're now spamming the comments. Someone will surely find it "offensive" or whatever 😜