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

Originally published at:




I think everyone’s missing the point. This fills the major market for creepy bad videos that go nowhere.

Alan Smithee


Who’s “everyone?”


It’s pretty clear that the real use cases for this kind of generative “AI” in creative work are pretty limited - highly specific contexts (e.g. “smart fills,” etc.*) that mostly function to take some of the busy work out of a job being done by a human being. The fantasy - of generating images and videos whole cloth - only works if you want to create “something,” but don’t actually care about the specifics. That’s fine for click bait images, where AI pictures are replacing random, cryptic stock photographs, or scams, or for would-be amateur art directors who don’t know enough to realize what’s missing when they exclude the artist who is considering how the image is going to be used, and what it means. But the “promise” is to produce usable work without paying artists and designers - the real uses don’t justify the hype, the enormous amount of money spent to get here, nor the resources required to do anything.

*I see research every week that uses neural networks to power some cool creative tool - but you’d have to have some awareness of what’s required in the relevant jobs to find it exciting (“Wow, we can extract much more usable motion tracking data from video without even using mocap markers!”). Pretty sure that’s the actual future of the tech and how it’s going to be used long term, however, after the dusts settles, the hype dies down, and a bunch of “AI” companies collapse.

I also found Ed’s other posts about the Rabbit “AI” device, specifically the stuff about the cofounder, to be interesting. He apparently went straight from an NFT game scam to this “AI” device scam - you can really see how people are trying to profit by riding whatever the current waves of hype are before they crash into the ground…

Indeed, who could have guessed?

(That seemed the perfect gif, for this context.)

Everyone who cares about quality.


the Shy Kids team produced Air Head with a team of just 3 people in around 1.5 to 2 weeks

That’s…not impressive to me? Then again, I’m not a filmmaker. I would love to see a team like Corridor try and produce a shot-for-shot remake, to see how much more work it would be to film everything conventionally, replacing the actor’s head with a (likely way more consistent) balloon using the tools of their trade. I suspect AI isn’t really saving anyone anything, given the mediocre quality of its output, even after extensive post-production.


SORA, and tools like it, are in their infancy. The creators haven’t yet addressed these features because they have been busy just getting the software to create something that looked passable in some contexts. You can bet they are hard at work addressing these issues, because they really believe this will be a useful tool for allowing “anyone” to generate a quality video.

Yes, this will result in a huge output of dreck for quite a while, and it will be some time before the software is good enough, and people are talented enough to prompt it sufficiently, to create anything that would be considered good by most standards, let alone take significantly less time and money than producing a movie today. It will be even longer before its creations are felt to have personality, which will always require at least one human to be in the loop.


Yes, I just watched some new AI video generator that Google is developing…the example was “cowboy riding horse on the plains” but the horses legs were all over the place as the AI struggled with the foreshortening. It was more like a giraffe, really.


Electricity sucks and has no value except shocking people you don’t like.
… is what this entire article sounds like.

You are worried about our jobs? Electricity’s gonna take our jobs!

LLMs are a black box discovery. At the absolute very least, they will change how humans communicate with software. I’m really not reaching very far. Sora is just the generation 0.001 pushing it a little harder to see how it handles video.

Your complaints on its value and sophistication won’t age well, probably not even in a year’s time, but certainly afterwards. Things are just getting started.

1 Like

I’m fairly sure your “AI” boosters will age much worse and much quicker.


It’s like we haven’t heard this exact same shit over and over again. Many times.

The actual breakthrough was in 2012 on applying brute force computing and it took the hardware a few years to catch up, and the production to meet demand, ever since then all they’ve been doing is inventing new gold rushes for their new fangled dynamite to use up.


Yeah, people keep selling these as something that will somehow keep get infinitely better. And yet right now the only plan groups like OpenAI have to improve their LLMs is brute force, and they have been saying they will need vast amounts more energy for it. That means their plan is for marginal improvements at great cost, and the fundamental limitations discussed – like that LLMs don’t actually have any internal understanding that allows them to make consistent scenes – are not being addressed. Everything about how it will inevitably overcome them is faith that people who actually understand the algorithms as more than magic black boxes tend not to share.


yeah, but, how’s the craft service?


episode 8 bullshit GIF by RuPaul's Drag Race

And yes, there are actual real concerns about technologies replacing jobs, because there is a long history of capital attempting to use technology to replace jobs. :woman_shrugging: You could try reading a book instead of just accepting whatever the rich and powerful say, maybe…

Seth Meyers Idk GIF by Late Night with Seth Meyers


I genuinely don’t understand this metaphor. Are you saying that I’m using AI to hurt people while also complaining about its overall existence? Are you saying that electricity was originally designed to replace workers and increase shareholder profits without actually adding anything of value to the acts of labor and production? Or are you saying that the idea of having a lightbulb in every home is the same thing as Hollywood producers using AI to replace labor?


It’s the kind of metaphor you use when you don’t understand what you’re talking about.


… and thus Sora 1.0 might be predicted to suck up 1000× as much energy, water, rare earths, etc. :thinking:


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!

(Dismantling Jupiter for Sora 3.0 is going to be harder, but we’ll burn that bridge when we come to it.)


The Luddites were right.

The Luddites weren’t against technology, they were against the use of technology by the 1% to take away freedoms, rights, and conditions from everyone else. They weren’t against the mechanised loom, they were against the loom being used to crush society and workers’ rights. (And, quite often, literally crush the workers themselves.) And that’s exactly what they were being used for.

The Industrial Revolution cannot be disassociated from Dickensien conditions: from the systematic dispossession and exploitation of huge swathes of people.

The Luddites were turned into caricatures of ignorant reactionism because the people who owned the newspapers and who wrote the history books were friends with the people who owned the factories and the tenements and the workhouses.

Praise the machine as much as you like (and there are aspects of the technology which are indeed worthy of respect from an engineering point of view) the people who own the mills are not going to invite you to join their club. You’re going to suffer in squalor down here with the rest of us. Because the people funding this aren’t trying to “change how humans communicate with software” for humanity’s benefit, just their own, and if they get their way, then will take another hundred years to claw back the lost rights and freedoms, some of which we barely have words for, because we never knew they could be alienable.