william_kent
Well-known member
that spring we lived on bowls of risotto | in the style of James Ensor
post pictures your computer has generated ( algorithmic art )
that spring we lived on bowls of risotto | in the style of James Ensor
Clinamenic
Binary & Tweed
Clinamenic
Binary & Tweed
I don't know, I think there is more to it than that. It reminds me, although in a more warped and alien way, of how a child's depiction of a scene is unrealistic and maybe impressionistic in ways that get kinked out as they conform their style to what they learn, and as they hone their craft, etc.
More precise tagging of images, presumably, could result in even more accurate "feeling" images.
Clinamenic
Binary & Tweed
Because these aren't truly randomly generated, there is a layer of human-legible order on the end of the training set, which is drawn from albeit imperfectly. I'm doing one with "star" "moon" and "wood" and all three make a discernible impact on the resulting images.
Clinamenic
Binary & Tweed
Actually I may have just been inconsistent with the input texts. I'm starting the round again with the same seed, and it looks to be identical to the last round.
I would be surprised if two rounds were totally identical but that's good to know you can get something consistent if you use the same seed
sourceful.us
Introduction to VQGAN+CLIP - 🟧heystacks
Here is a tutorial on how to operate VQGAN+CLIP by Katherine Crowson! No coding knowledge necessary. - 🟧heystacks
Clinamenic
Binary & Tweed
Here is one with "wood | moon | star" that I took some of the images from and rearranged them into this palindromic sequence.
And here, from the same GAN results, but a different selection therefrom, and in a different order.
And here, from the same GAN results, but a different selection therefrom, and in a different order.
Clinamenic
Binary & Tweed
I want to also establish some more techniques here. I wonder if I input "brown top | cyan bottom" if it will separate the colors into the top and bottom halves of the image.
Clinamenic
Binary & Tweed
Clinamenic
Binary & Tweed
Corporate feudalism with AI CEOs. Pay your data fealty!
Clinamenic
Binary & Tweed
Kidding, of course. Still a faithful utopian.
Clinamenic
Binary & Tweed
This mentions how computationally intensive these neural networks are, which I was worried about.I would be surprised if two rounds were totally identical but that's good to know you can get something consistent if you use the same seed
Introduction to VQGAN+CLIP - 🟧heystacks
Here is a tutorial on how to operate VQGAN+CLIP by Katherine Crowson! No coding knowledge necessary. - 🟧heystacks
One could be thrifty in how these aesthetic experiments are handled, in the effort of forming a kind of pidgin interfacing technique with these things.
The "brown top | cyan bottom" input would be a simple probe, to see how this GAN interprets spatial/directional inputs.
And beyond just changing the parameters, one can manipulate and reconstruct the results.
@catalog @william_kent did one of us mentioned the possibility of using image files as input? That may allow a whole other panel of levers to pull.
But at this point maybe one should just learn to code them directly, and I am only starting to appreciate how energy-intensive this whole practice is.
luka
Well-known member
what is it you like about it? the wibbling?
@Clinamenic Yeah you can upload an image, use the Spanish language one.
you need to put it in the temporary folder to the left of the parameters area, then copy the file path to it and enter it into thd initial image field. On my phone atm so can't really do a screenshot easily but if you can't figure it out, lmk
you need to put it in the temporary folder to the left of the parameters area, then copy the file path to it and enter it into thd initial image field. On my phone atm so can't really do a screenshot easily but if you can't figure it out, lmk
william_kent
Well-known member
The GAN I use allows the use of an array of images which can be referenced by URL - e.g., ['image00', 'image01', 'image02',] - it's also possible to do an array of strings, like ['this is phrase one | in the style of something', 'this is phrase two | in the style of something else' ]..
Combining arrays of image and text prompts can be interesting.
Last edited:
william_kent
Well-known member
@Clinamenic last night I had a thought that "Toward a Mythos of Extropy" would make a great album title for a prog rock or metal band and went ahead and made the GAN generate some fake LP covers..
Toward a Mythos of Extropy | progressive rock album by 'Clinamenic' | in the style of Roger Dean
Toward a Mythos of Extropy | symphonic black metal album by 'Clinamenic' | in the style of symphonic black metal album art
seed: 0.666
Toward a Mythos of Extropy | progressive rock album by 'Clinamenic' | in the style of Roger Dean
Toward a Mythos of Extropy | symphonic black metal album by 'Clinamenic' | in the style of symphonic black metal album art
seed: 0.666
Clinamenic
Binary & Tweed
Really my fascination is philosophic mainly, but also pseudo-scientific. We as humans have our parameters of perception (the visible and audible spectra, for example) as do these neural networks. Only with the NNs we get to completely program them.
Clinamenic
Binary & Tweed
@catalog @william_kent is this something that, with greater programming literacy, one could run on their own GPUs? Really I have no idea how simple or complicated that would be.
Clinamenic
Binary & Tweed
Can electroencephalographs be translated into binary to serve as some kind of custom "noise sheet" for these GAN iterations?
Clinamenic
Binary & Tweed
I envision myself seated in an antique leather chair, scalp adorned with electrodes, surrounded by racks of GPUs.