Stable Diffusion makes videos now?
(twitter.com)
You're viewing a single comment thread. View all comments, or full comment thread.
Comments (14)
sorted by:
This stuff is still super hardware/power intensive. It take can 6-24 hours to render a movie like that, and you're leaving a LOT up to chance.
And it still is bad at doing anything with strong coherence. It can make pieces that look technically impressive but they lack composition. If there's any kind of context or meaning in the pic it's a fluke.
That's what I think every time I see some 'modern art'. Fitting.
Lot of modern art is money laundering bullshit. Get away from the heavily promoted shit and you'll find some quality stuff out there.
The ONLY way I found to have any coherency is making textual inversions and even there you occasionally have some weird issues. I think I fucked myself when I picked a character that has got a hairpin on one side of her face, I flipped the images to have more training and now she just occasionally has 2 hairpins on both sides, lol. Even if you work on coherency you're still having issues like that. I'd expect this to be worked on within a couple years, though. I expect something better to happen in terms of textual inversion also.
especially the learning aspect of AI is super hardware intensive, let alone make a 30 frames movie which is 30 images it has to generate for 1 second...but I've not done a lot of animation with AI(yet).