Comment on Adobe's new generative AI tools for video are absolutely terrifying
RGB3x3@lemmy.world 2 months ago
“Absolutely Terrifying”
Really? Did he watch the same video I saw on that page? That b-roll was really bad.
I get that it’ll only get better, but generative AI models will never understand filmmaking techniques because it can’t. That’s not how these models are built.
It can create a city skyline and pan right, but it’ll never know why a pan right in that scene was appropriate or how the lighting fits in with the rest of the scene. It’ll never come up with new ways of “filming” a scene, because it’s all built on what already exists. There’s no style to generative AI.
Yes. It is a new tool for vfx artists and not a replacement. If they can deliver higher quality for less money, you’d expect them to be more in demand.
“Never” is a big word, but it’s really not clear how one would train an AI to know what it should generate. See the hubbub about diversity in google’s image generator. I see no theoretical problems, but in practice it’s just not going to happen any time soon.
VirtualOdour@sh.itjust.works 2 months ago
You say that like the current models are the end of the line but understanding why film making techniques are used isn’t impossible even just for a llm based system. Designing new styles isn’t out of reach for ai either, sure you can God of the gaps it and say there’s a mysterious sliver of soul required but practically it’ll be able to be every bit as original as any human, probably more so as it has more knowledge to work from.
I know it’s desirable to hate on ai because it’s scary or popular but nailing your colors to the argument that it’ll never be able to do certain things is already an exhausting game of moving that goalpost every time a new model emerges and that’s only going to continue.
technocrit@lemmy.dbzer0.com 2 months ago
It’ll be so “original” that it makes no sense, evokes no emotion, and goes nowhere.
hikaru755@feddit.de 2 months ago
Sure it’s pretty far away, but it’s also moving at break neck speed. Last year low-res spaghetti-eating Will Smith body horror was the pinnacle of ai generated video, today we’re already generating videos that take at least a second look to determine that it was AI generated. The big question is at what point that improvement rate will start to level off.
VirtualOdour@sh.itjust.works 2 months ago
It’s not really about the ai making a complete movie though it’s about emerging tools allowing creators to make their vision.
But yes they’ll be making complete movies every bit as enjoyable as marvel sooner than you can imagine. And movies with emotion and message that speak to people on deep levels, that’s really not as hard as you might want to think.
And yes we do, making movies is just an easy to play with toy the real work is going to understand these tools so we can use them to do things like generative design of 3d printable items to set metrics so you can just say ‘my dishwasher broke’ and it’ll look at the photo, analyze the problem, design a replacement and offer fabrication options. That’s the sort of thing this leads to, being able to have repairable and upgradable hardware, being able to customize your life and live better for cheaper.
It’s a world where the impoverished that used to toil in sweatshops so we could have cheap clothes can get world class education in their first language with questions answered and conceps demonstrated using examples they can understand and with clear and well constructed visual aids. Allowing them to create the things they need to improve their lives and recreate the local sustainability we tore from them.
If this allows someone with a fascinating life story and perspective to express themselves and create their vision then it’s a great thing, I’m looking forward to seeing random movies from obscure areas of the world and I won’t miss corporate and polished made by the numbers Hollywood the slightest bit.