The following images were made using Leonardo.ai, my build of Disco Diffusion, photo-manipulation, photobashing & sketching in Gimp and Inkscape, and post editing/composition in darktable.
Back to the roots! I generated a few portraits using Deforum and Kadinsky, that I used as an initial image cut into 7 slices into Disco Diffusion. Here is the original initial portrait
Once this guy has been cut in 7 parts (Top left & right, bottom left & right, center top & center & bottom), I fed it into my custom Disco Diffusion to create detailed slices (slices + settings following)
ViTB32
ViTL14
RN50x4
Ddim
No secondary model
250 steps
Guidance scale: 77000
Tv : 150
range : 150
Sat : 2500
Cutn batches : 12
Initial image generated with Stable Diffusion
Eta : 0.85
Clamp : 0.095
Custom cutn schedule
And, all these slices finally assembled back together, after heaps of photomanipulation and adjustments, edits in Gimp and darktable, lead to this following result :
The idea is here, but the madness is a bit "too much", the incoherences are a bit over the top.
So, I decided to feed THIS as an intial image into Leonardo AI, with very high initial image scale impact, and after playing around a bit with settings, it ended up looking like this, which I find very cool
