If you appreciate what I’m doing feel free to https://www.buymeacoffee.com/Uzia, buzz the life outta me or share some of your generations! I love to see them.

(On that note though, imagery depicting sexual violence/abuse will be reported immediately.)

Ok, this one came fast. V2 left me feeling unsatisfied with is take on art styles so I decided to do a reshuffle but also added Leosams Helloworld into the mix (link below). This was originally going to be a 2.something release, but rather than just fixing the issues with the art styles, the overall improvement deserved a full version. Composition, poses, expressions, details; all see a step up in V3.

https://civitai.com/models/43977/leosams-helloworld-xl

(Clears throat. Takes a deep breath. Summons the most ******** action hero movie trailer narrator voice you’ve ever heard in your puny armed life).

“PREPARE TO UNLEASH YOUR CREATIVITAH!! This! One of a kind, groundbreaking, face crunching S. D. X. L…” Coughs uncontrollably, “Sorry, had a lump of grandeur stuck in my throat.”

Soo. V2. First off, I personally harmed no Loras in the making of this merge. I do so solemnly swear on the almighty data center. That said the changes to non-photographic styles in V2 is at times substantial. Skin details are better although at times the pores look like they could poke yer eye out. And I have the feeling it follows my prompting better.

All generation parameters stated for V1 apply here too although I’ve been rekindling my relationship with that raunchy beast known as, DPM++ 2M Karras.

Bonus Gallery:

There is a preview image gallery on this page on this page with no prompts. That is because the images were made in Fooocus which doesn't save the meta data. But, the point is they are examples of using Opiate.v.4 or Lysergic.v3 as a refiner:

https://civitai.com/models/69587/opiate

https://civitai.com/models/63628

Values that work seem to range from 0.25 to 0.8.

Bro Science

The ingredients here are TFV version one. Then:

https://civitai.com/models/133005/juggernaut-xl (Latest version)

https://civitai.com/models/255907?modelVersionId=288506 (Restoring balance to the waiforce)

https://civitai.com/models/133808?modelVersionId=324958 (Bringing said balance round the back with a shotgun.)

https://civitai.com/models/259280?modelVersionId=303498 (Waiting in the garden with a spade)

As far as process goes, nuthin fancy. Just making and blending lots of different merges until one of them became my go to model. Hope you guys enjoy.

V1 (And generation tips)

SOOOooooo. I've been chiseling away at SDXL for the last few months trying to find the balance between art styles. This was originally going to be OpiateXL; the original idea being I'd only use models that I was sure had no loras baked into them. A kind of strong hold for the work going into fine tuning atm. But, when taking into account the stated restrictions on certain models, what I was left to work with could be counted on one hand. The thing with SDXL, if you want to try and find a balance between all the different styles it can do, well, it's just too fucking vast. I have not added a single lora into this merge. There was however, a war. A war as old as time. The war between saintly artist integrity, and boo... I mean freedom of expression. OpiateXL exists but in the end it just became the seed that would become Too Fucking Vast.

(Unexpected shout out to NobodyButMeow for both the review and submitting a bug report to get onsite generation up and running, had no idea it was bugging out, cheers.)

Usage Tips:

Samplers:

Euler a: I like to use this mostly for cgi styled images.
DPM++ 3M SDE Karras: General purpose workhorse, I like to use it for pretty much everything.
UniPC: Been experimenting with this one for photographic generations. Looses definition and quality in exchange for more, or, a different brand of realism. I think I like it, jury's still out.

Cfg:

4: Great for pretty much everything but especially photographic styles. (My favorite)
5 -> 7: Great for pretty much everything, as it goes higher it's use for photography diminishes but is as, if not more, useful for artistic styles.
8: Great for some art styles but will burn photographic generations to a cinder.
9 and above: Yo crazy.

Steps: 20 to the limits of your patients. I never really use more than 50 and find that above that it's a case of diminishing returns.

HiResFix:

(Standard use, though HiResFix in general is optional and can add quite bit of time to the generation)

4xUltraSharp
Denoising Strength: 4
Steps: I tend to use around 50% of original step count.
Upscale: I find 1.5 to be enough with SDXL.

(Fancy use)

4xUltraSharp
Denoising & Steps and their sometimes horrid relationship: The higher you go the more unstable it is which can conjure horrors I could have spent my life time without seeing, especially with photorealic prompts. But. For crazy trippy illustration stuff you can hit a Denoising of 7, with 40 HirRes steps, but also 40 normal steps. This can add a freakish amount of detail, so if that's what you're after give it a try.
Upscale: Same as above but you do you.

Refiner: Na, I'm good.

Useful Resources:

Artists... Maybe even all of em. This page is feckin epic and provides example generations based of the Artists name, so even if it's not their style in real life you can see what SDXL does with their name. Of course it varies with each model but to get an idea of what we can do, now that we're standing on the shoulder of these giants, check this page out: https://weirdwonderfulai.art/resources/stable-diffusion-xl-sdxl-artist-study/
Resolutions from the interwebs, says it's for finetuning but I use it for generation too:

Bro Science:

Step1: Make an incorruptible, pure of heart, squeaky clean and (presumably) Lora free model that hits the way I want it to. Made with...

Crystal Clear XL (Thanks for permission)

https://civitai.com/models/122822?modelVersionId=133832

Juggernaut XL, (v3 and v5)

https://civitai.com/models/133005/juggernaut-xl?modelVersionId=198530

Run Diffusion XL

https://civitai.com/models/120964?modelVersionId=131579

ThinkDiffusionXL

https://civitai.com/models/169868/thinkdiffusionxl?modelVersionId=190908

Step2: Get corrupted. Make a second model for boo... artistic freedom. Made with...

Better than words

https://civitai.com/models/149359/better-than-words?modelVersionId=172253

TalmendoXL

https://civitai.com/models/119202/talmendoxl-sdxl-uncensored-full-model?modelVersionId=131960

Venus XL

https://civitai.com/models/140329?modelVersionId=157315

Pyro's SD Models - Blowjob

https://civitai.com/models/126169?modelVersionId=137902

Step3: Supermerger. Make and mix, create and kick the crap out of different candidates. Make some terrible models that go with other terrible models to make a better model, see how it feels, delete it, start again. Smoothstep the skin from what was OpiateXL and put it onto the bones of, lets call it CorruptionXL, in an effort to pull back some of the art styles from the school of hungry loras swimming the corrupted waters. And then merge and test until I have something that I like.

Thus ends the tale of TFV's birth. And, did I succeed? I think so. Kinda. You tell me. I had a lot of fun making it. And I hope you have fun using it.

TFV.SDXL.BAKED - TFV.SDXLv1.0