Strength anywhere from 0.4-1.1. Usually, 0.7-0.9 is the sweet spot for effect and depth. Seems around 25-30 steps does the trick pending sampler. Optional trigger is I84c1d4u_c0l0r_sh1ft. Does something sometimes. Seems to boost depth a ton if used toward the end of the prompt.
For the Walter Potter taxidermy stuff, use anthropomorphic {animal of choice} and then talk about it as if it were human. I reserved the word taxidermy for regular old taxidermy!
What is this?
What if everything in the world was actually colored like autochrome and old colorized photos but not photos (unless you want)? Extreme hues, lots of drab aside for a random color or two extremely oversaturated. Make it extremely 3D and capable of some wicked depth. Nice. Why not add some taxidermy on command too? VERY EXCITING. Specializes in long, cavernous, busy, and surreal objects. This makes no sense. Just see the pictures.
Some Training/Inspiration Ramblings
So, this was born out of noticing lots of little things with Flux LoRA training over the last year with artifacts, bokeh backgrounds, and depth (wrestling ring ropes, incidentally, will drive you mad). I thought I could leverage some of that knowledge and came up with the data set for this with the right object composition (and colors- that is the easy part though).
I took a different approach with this one…. think about image diffusing from a perspective of “what can I see in the clouds/can I play Jackson Pollack connect the dots/what is your crappy wireframe model supposed to be” as steps form an image. Something like say, a cave, would also have the same mid-diffusion form, to an extent, as the tube of a waterslide, a bathtub to a shrunken person who lives in the drain, or even the inside of intestines in an image from a colonoscopy. Leverage that with super high-res upscaled images and things happen 😊