Diffusion Walks: Series 1

first of all, zoom in. then zoom out. more.

implemented a modified "Image Random Circular Walker"
to use sharp, rotating lines and jagged spirals instead of circular lines
by breaking radius/direction/angle a little bit, increasing stroke speed,
and utilizing my own finetuned diffusion model outputs as inits for the new walking algorithm.


The new walk is some what of a frantic spidercrawl, accompanied by a shimmering dance.
I enjoy the glints of frenzy that this sort of motion adds to otherwise serene diffusion outputs.
Furthermore, there's something special about the output being revealed; re-distorting a guided
set of denoised images gives a lot of depth and momentum to the work for me personally.
It turns the process into a cycle.

fxrand() affects the init image, stroke speed, number of strokes, duration, stroke angle, and stroke sequence direction.

there are 25 image variations.
this is not about the images as much as it is about the walk.
there are countless walk variations.

duration, stroke angle and direction ranges vary as well.

there are 6 potential speeds.

as a result, hash variations will affect clarity, palette, init, and pattern of the token.

no recorded hash-unique features/attributes are printed on the token as a guide
please experience each variation's infinte iterations as their own realm
explore the variations, find different speeds and walker paths, enjoy the boundless variety in this set


zoom out to see the full image. zoom in for strokes. open live for easier zoom experience.
1280px by 1664px.

credits are important. when you play with a lot of things you made
mix them with things your friends made
and break things other people made
it is paramount to give all parties their flowers once
the art is made.

generated with my finetuned SD model, deforum notebook, and p5.js
credit to Vamoss for the original image walker algorithm (https://openprocessing.org/sketch/624879)
which was the basis for the new walker algo
credit to Stable-Diffusion team for original model which I tuned
credit to deforum team (which includes myself) for the deforum notebook

License included in License.md and p5.js file
License for Image Random Circular Walker and Diffusion Walks are both CC BY-SA 3.0