Level4 SDXL (Turbo)
A reimagination of Level4 for SDXL! Currently in alpha and improvements are on the way. Face and hands still need work.
Update:
Level4 XL alpha v0.4 brings higher contrast along with better hands and faces. This came at a small sacrifice to flexibility. Some styles may need to be weighted to reflect this, e.g. (style:1.5). AutoCFG is not required but highly encouraged. This model is sensitive to CFG changes. Different styles may require different CFG, which is why I suggest AutoCFG.
Recommended Settings:
Steps: 8-16
AutoCFG: 2-3
CFG: 1-3
Sampler: DMP++ SDE
Scheduler: Karras
Contact me if you wish to use my model commercially!
Description
Level4 XL alpha v0.4 brings higher contrast along with better hands and faces. This came at a small sacrifice to flexibility. Some styles may need to be weighted to reflect this, e.g. (style:1.5). AutoCFG is not required but highly encouraged. This model is sensitive to CFG changes. Different styles may require different CFG, which is why I suggest AutoCFG.
FAQ
Comments (11)
Don't really understand why this is not hyped more. I'm as excited as I was when I stumbled on an unknown model named Juggernaut. Finally I feel the pictures have a Midjourney-style depth and density! Can't wait for the development...
I'll come back when there's a XL 1.0 version, looks neat!
Great model, but somehow the description or the example images are wrong: If I use CFG Scale 3, like in the example images, I get very different images (in A1111), but nearly the same, if I reduce the CFG to 1.
For example "woman waving at viewer" gives in A1111 with CFG 3 indeed an "80's anime screencap" as requested in the prompt, but with CFG 1 it gives a similar analogue photo like the example. (?)
So the actual prompt adherence is actually much better(!) than shown in the examples.
PLEASE RELEASE A NORMAL VERSION
fantastic model, it seem to ignore wide shot, zoom out, landscape, full shot, medium close up, and force close ups 90% of the time.
otherwise what a amazing model. excellent work thank you very very much
Anyone had success with this model using Automatic1111? I try the recommended range of settings (low CFG, steps) but everywhing comes out rough and super-saturated. TIA for any suggestions.
Wow, you can really be proud of your model. It is really awesome ! It's super fast and give amazing and interesting results. Many Thanks !!!
Amazing results, this model kicks some serious ass. Please keep developing it !
I'm looking forward to your next model <3
Great Model! If now it were to actually do some shots that are not portrait-esque, but a bit further away (full body view), it would be even more amazing. Edit: Somehow, this is the only Checkpoint that does not play nice with any HiRes Fix Upscalers whatsoever.
Love the model. What's a good inpaint model to use with it?
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.