Random Musing

[malymin]

If you want to develop an art style that is hard for AI to imitate...

You might want to consider pixel art?

It's possible that this weakness will get fixed as the models develop, but for now - image generation seems to be consistently dogshit at pixel art. People ask it to make pixel art and it makes, like... pseudo-pixel art. There's mixels, there's jpeg-like artifacting, it looks bad. Mimicking photography, digital painting, and vector-like flat shaded art all seem to be easier for the models than pixel art - probably because the training data contains more of those things than it does pixel art, but I also wonder if there's something intrinsic to how image synthesis works that makes understanding "pixel art" as something more specific than "theres a bunch of squares and rectangles" difficult.

This also reminds me of the issue of how hosting images is one of social media's sources of bandwidth issues. So much of current fanart culture is focused on posting high-fidelity images with large filesizes, but ultimately every website with unlimited image hosting is going to run into the server and bandwidth costs involved. Was talking about how this bodes badly for the future of conventional social media that doesn't plan on using predatory practices (selling personal data, etc), and how the internet at large might need to pivot back to a text-centric culture if we want more socmed options that are both non-exploitative in their monetization models and financially sustainable (but most people i've mentioned this are afraid this would mean the death of fanart in fandom), they mentioned:

my first thought is getting back into the art of making small-scale pixel art and such, that kind of thing used to be all over the internet and generally the filesizes for them are a lot smaller than large-scale renders and digital paintings and such

I feel like art optimized for small filesizes is something that's become... less of a common skill? I miss tiny-handful-of-KB pixel graphics being all over the place, but also typical social media displays cute little animated gif pixel arts... abysmally, because their imageposting format is designed for photosets. My conversation partner added:

yeah true, display sizes on social media aren't geared towards smaller-scale images i wonder if the "add gif" kind of feature that some sites have could be tuned to account for small pixel and reaction kind of gifs to display them more clearly, i feel like it'd be relatively easy to make that a popular feature considering how much people like using emojis

I think about how decorative pixel graphics used to be very popular on Tumblr back when users were encouraged to customize their page's HTML (remember those Homestuck "pixel families" and the edits of them to represent other fandoms?), but I've noticed that Tumblr has been... not outright removing the ability to make custom HTML pages, but making it harder for new users to even know it's an option, and altering the UI to discourage viewing the site via custom pages and encourage viewing blogs the more standardized pseudo-mobile format.

yeah, customization being removed on a lot of bigger sites definitely seems like a factor in stuff like that declining

I think there's multiple good reasons to shift back towards pixel/sprite art in the wake of fandom artist anxieties over being replaced by AI, the decay and collapse of last decade's Standard Online Experience, etc. But that doesn't mean people are actually going to try, or even think to do it. And I'll never be persuasive enough to start any kind of movement, so.