2.0 and 3.5 version of my mix merged with offset noise model
2.0 = better at realism
3.5 = better at fantasy
Description
FAQ
Comments (21)
Great mixes, both !
Can you add, if possible, a smaller 2gb pruned fp16 no-vae safetensor version ?
.ckpt version plz
Sure gimme 30 minutes
i will upload ckpt por the fp16 versions
Done!
Great results. Completely solved the problems I was having with the non-offset model, and feels like there are other improvements too.
One difficulty I had was that in overall dark images there seems to be a hair trigger between monochrome and hyper-saturation. Not sure what to prompt to retain color without it flipping to extreme saturation, things like 'muted colors' don't have a strong effect.
@sooperl did you tried negative prompt : (monochrome:1.3), (oversaturated:1.3) ?
I did try those kinds of prompts, and what I mean is hair trigger is that it still tended to images that were either saturated or monochrome, and it was more like the ratio was adjusted, though there are exceptions and some proportion of the images do have the intended saturation profile. It means the discard ratio is greatly increased when dealing with very dark images. When not creating high contrast low-key images, this is just great model and now my go to choice for anything that isn't a person. Even though it has some awkwardness around the dark images, it's still the best option by far.
looks nice!!
I'm not even using this for its intended purpose, and it still delivers amazing things. It's producing all kinds of interesting urban images, awesome water, and intricate industrial detail.
How do you achieve those urban images, I'm interested in using them in my architecture study
There's alot of differences between these sets of images A/B respectively. What is the difference here?
Is this 2 different versions you're showing and if so which is which please? The gen data does not show the model so I can't tell what I'm looking at
wdym with A/B?, this set of images are the combination of my original model with the offset model thats all
@cheesedaddy He means which image is which? Is the left the original or is the right?
@BobRoberts they are 2 images with the same prompt but different seed
Sorry, new to all this so forgive my naive question: If I were to create a fantasy scene using this checkpoint, how would I then put a couple figures - like elves on a hunting expedition - into the scene?
I suggest you use the tile plugin to achieve the desired effect
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.