OpenAI's generative video software is more tedious than actual filmmaking

[Frank_Butty]

The fact that they even could produce anything coherent is the win to me. I expect the quality of both inputs (via experience) and outputs (via technology) to improve significantly in near term making these complaints moot. Is it art? No, certainly not. But neither is 99% of the “creative” excrement we generate ourselves

[anon61221983]

@Frank_Butty

But neither is 99% of the “creative” excrement we generate ourselves

[chenille]

As mentioned the companies have no ideas for how to make significant improvements, just throwing orders of magnitude more resources in and hoping. As for the things humans make themselves, even badly done art sometimes has interesting ideas or aspects to it. “AI” is always the most generic version of the queue by design.

[Brainspore]

@Frank_Butty

I expect the quality of both inputs (via experience) and outputs (via technology) to improve significantly in near term making these complaints moot.

A computer algorithm cannot learn and improve from “experience” as a human can, because a computer algorthith doesn’t have “experiences.” All it has is a buttload of sample data that it doesn’t actually comprehend in any meaningful way.

Artificial Intelligence is kind of like artificial turf. It’s better than it used to be, impressive in its own way and maybe even functionally superior for some extremely narrow use cases. But no matter how good the technology gets it is a fundamentally different thing from real grass and will never make a very good general substitute.

[anon61221983]

Also, what’s “good” art is always, always subjective… what some find wonderful art, others find atrocious… but art is about what speaks to you, not about some kind of objective aesthetic criteria…

[chenille]

That’s a good point. I meant bad art in the sense not of taste but skill at realizing your ideas. Everyone starts as a beginner. I’m one myself but it hasn’t stopped me from making things other people like.

[Melizmatic]

@Frank_Butty

we generate

[anon61221983]

@chenille

I meant bad art in the sense not of taste but skill at realizing your ideas.

Oh yeah… I just meant that what one considers “good” is subjective… both our points are true, I think.

[KathyPartdeux]

I’m looking forward to them replacing techbros and billionaires.

[chenille]

[smulder]

[gatto]

@LurksNoMore

Not to worry! By dismantling the inner planets and turning them into a Dyson swarm of computronium and solar panels, we can run Sora 2.0 with no problems!

twice in one week!

The 11-foot8+8 bridge has met an unstoppable force (video)

but i need the dyson sphere to power my bitscoin!

[vermes82]

OpenAI's Video Generating AI Is Dead On Arrival - Better Offline | iHeart

Here’s the audio version of Ed’s piece.

[GagHalfrunt]

Sora can’t carry anything over from one shot to another. Fixing that would require fundamentally changing how AI image generation works.

@thomdunn

The very nature of filmmaking is taking different shots of the same thing, something that I anticipated Sora would be incapable of doing as each shot is generated fresh, as Sora itself (much like all generative AI) does not “know” anything. When one asks for a man with a yellow balloon as his head, Sora must then look over the parameters spawned during its training process and create an output, guessing what a man looks like, what a balloon looks like, what color yellow is, and so on.

It repeats this process for each shot, with each “man with balloon as his head” character subtly (or not-so-subtly) different with each repetition, forcing users to pick the outputs that are the most consistent. Achieving a perfect like-for-like copy isn’t guaranteed, but rather, filmmakers must pick the ones that are “good enough.”

[Kilkrazy]

At the very least, you can talk with a human artist, ask them about their influences, criticise their ideas, compare them with your own situation, and set them within a social milieu. However valid that may be, you’re having an interaction with a person.

An LLM or similar, merely calculates that a number of pixels, with a variety of RGB values, is 83% likely to be adjacent to a slightly different number of pixels and RGB values.