Comment on Terry Gilliam says he doesn't have enough money to make ‘The Carnival at the End of Days'
mindbleach@sh.itjust.works 2 months ago
He never had enough money to make any of his movies.
In total sincerity - this is what video generators will be for. Letting the robot make it all up from one sentence is an overblown demo. Anyone with a killer script and a great visual imagination should be able to make a two-hour film instead of a comic book.
The technology modifies existing frames. If you feed in some actors doing a scene, and describe what it’s supposed to look like, you should get out dirt-cheap CGI to fix up all the details. The closer the input looks, when blurred, the better the results should be. So if the script says the clouds part and a giant hand gives us the finger, well, grab some pillows and a ladder.
Or, this being Terry fucking Gilliam, animate some flat cutouts, and have the computer make that visual punchline plausible.
Emperor@feddit.uk 2 months ago
So… “use AI”?
mindbleach@sh.itjust.works 2 months ago
If you want to oversimplify future expectations of a brand-new technology to two loaded words.
The people insisting it’ll never be useful are as off-base as the grifters insisting it’s the next everything. Right now, on consumer hardware, you can take a photograph of a toy spaceship and tell your video card to make it look real, and it will. No kidding we’re gonna see that used professionally.
Even if we’re still way too early - sometimes the alternative is that a thing does not get made. Fancy new ways of faking stuff don’t need to be flawless, when they’re competing with non-existence. Early special effects were laughable. Early CGI was hideous. But they did things a production couldn’t possibly do, or couldn’t afford to do. If this stew of linear algebra and pirated DVDs lets weirdos like Gilliam show us a mildly gloopy version of what’s going on his head, with a budget he can scrounge together in less than thirty goddamn years - great. I’d rather have a mildly gloopy movie than what-ifs.
Tippon@lemmy.dbzer0.com 2 months ago
Out of curiosity, what can you use for that? I haven’t kept up with AI, but that sounds like the sort of thing that my kid would love to try :)
mindbleach@sh.itjust.works 2 months ago
Stable Diffusion seems to be the done thing. Apparently you provide a list of tags with numeric weights, and it “denoises” all the ways the image doesn’t fit. From-scratch images are generated from a canvas full of random noise.
Basically “remove all the marble that doesn’t look like a statue.” The missile knows where it is because it knows where it is not.