mixed_biscuits
_________________________
Tragedy of the commons in action
Yeah I'm not sure if its the particular GAN itself or if its because higher amounts of text just confuse GANs more generally. But yeah I wasn't getting especially interesting imagery, just found it interesting that you can type in new prompts and it almost keeps up in real time.The output is impressionistic to say the least.
it's actually pretty decent, although no option to add a text prompt.hypnogram seems to have died - not working for me from home or at work - I wouldn't be surprised if they have run out of money for the GPU processing, it's been hammered the last few weeks - back to google collab for me
@catalog here's an online AI that allows you to upload photos - i've not investigated further because it relies on a "credit" system ( means money changes hands.. ) - it appears to be a variation of STYLEGAN ( i.e., style transfer )
The free google collab seems like it gives enough for our purposes. Pretty sure its just a daily refresh.not that i know of, but i feel like there's gonna be loads more web front ends coming up. paying for that processing power is the issue definitely.
it's actually pretty decent, although no option to add a text prompt.
used the initial image (on top) and then selected "bitcoin goddess" (below) (think that's what it's called) style
View attachment 9274
and it makes this
View attachment 9275
used all my credits doing that but you can buy more so might do.
this is great stuff - you can use the colab notebooks to create any resolution you like (with concurrent processing time required) but running those on a phone would be pretty difficult. i don't think impossible but wold be pretty annoying.Related to this thread, but with a little more human intervention; I use an old android phone that was kicking about the house doing nothing to generate GAN landscapes online, then feed them into a glitch app to fuck them up a bit. Delete bits... keep bits... build up layers of glitched artefacts, before finally processing with photo editing software, (again, limited to Android apps only).
My main problem with GAN generation is the low resolution output, suitable for viewing on screen only. With these I've been working to effectively obscure the low res GAN and upscale it so that the full size files are roughly an A4 600dpi resolution.
So its artwork produced via selection. Picking the best AI generated landscapes, picking the best automated glitch artefacts, all done with nothing more than a few setting sliders and an erase tool.
www.glitchedwork.com
You consider representing these as NFTs? You could do a limited series.Related to this thread, but with a little more human intervention; I use an old android phone that was kicking about the house doing nothing to generate GAN landscapes online, then feed them into a glitch app to fuck them up a bit. Delete bits... keep bits... build up layers of glitched artefacts, before finally processing with photo editing software, (again, limited to Android apps only).
My main problem with GAN generation is the low resolution output, suitable for viewing on screen only. With these I've been working to effectively obscure the low res GAN and upscale it so that the full size files are roughly an A4 600dpi resolution.
So its artwork produced via selection. Picking the best AI generated landscapes, picking the best automated glitch artefacts, all done with nothing more than a few setting sliders and an erase tool.
www.glitchedwork.com
I have some standards. Anyway, how would I compete with pixelart monkeys?You consider representing these as NFTs? You could do a limited series.