Open-Sora does pretty good video generation on consumer GPUs
backprop.coI know it’s probably using < 1000x compute of the real Sora, but “pretty good” is stretching it
Depends on your frame of reference. Compared to anything else I've seen generated on a consumer grade GPU, I'd say these are are indeed pretty good.
Here's their example gallery: https://hpcaitech.github.io/Open-Sora/
Compared to the outputs from other models run on consumer grade GPUs, I'd say those are very good.
Looking at it in low res (small rectangle within the webpage on mobile) they actually look great!
What's the useful frame of reference?
Looking better than other things that are also bad is sort interesting in that it represents progress in some direction, but it isn't very interesting to people outside of the topic.
For me I use the Will Smith video[0] from just over a year ago. Compared to the examples it's a pretty stark difference.
https://arstechnica.com/information-technology/2023/03/yes-v...
Yes, people learned not to generate other people eating. Current SOTA models still have no concept of walking (left leg, right leg, left leg, right leg; it's so complicated?), there is no reason to believe that they have learned the peculiarities of food consumption.
We seem to be in an exponential uptick phase of tech driven by hardware improvements; a few years ago this was impossible on consumer grade GPU. So in some sense there isn't a useful frame of reference, state of the art should improve out-of-sight about every 2 years and eventually I'd expect iPhones to be outgenerating Disney at movies.
Not GP, but when I looked at the examples, I thought that those already look pretty useable in comic book-like storytelling to set the mood. I.e. in settings where smaller details of the scene are not relevant and are not taking away from the "larger product".
Good that this frame of reference is hn and not some random website where people have no connection to ml...
Just run all key frames through stable diffusion and it should be quite good.
I don't think there's such a thing as key frames here, just frames. And if you run SD through every frame, the output will be janky because SD doesn't know about temporal coherence.
As soon as it's an MP4 it will have key frames all right. You could add AI upscaling to your encoder. People are making fun of "Just", but I believe I could take apart ffmpeg to add this feature (PoC) in two weeks or less. Provide somebody pays for my labor and for the HW.
Honestly neither does OpenSora it seems, as it is pretty damn janky already.
You can pass a few frames as a single image grid. Then you will get coherence, although it will be very limited by gpu ram.
>Just
Adding the word 'just' doesnt make it any easier. Something I've noticed is that people who have never done something themselves and are telling someone to do an difficult task, will use:
"Just"
in-front of it.
This is particularly relevant in tech.
Depends if it's already been done before, in which case "just" would then have been just used quite justly.
It's very likely the comment you replied to said it in a joking sense
It is extremely difficult to tell if the person is joking in a field full of people who think AI is some sort of magic.
Well I mean calling any of this diffusion/LLM stuff "AI" is a misnomer to begin with.
It's probably not the best idea to base your clone on the name that is being trademarked: https://trademarks.justia.com/984/05/sora-98405663.html
It looks like it's being developed by a Chinese group. What's OpenAI going to do, sue them in a Chinese court?
Take down their GIT Repository with perfectly valid DMCA Notice, I guess?
Does DMCA cover trademarks? I think it might only be for copyright.
They can probably sue GitHub then (their current hoster of their repo), the traditional way?
Though actually I imagine GH would just close the repo before it comes to that, because it looks like a pretty clear case.
I guess it's also possible OpenAI doesn't care about this.
Yeah I'd imagine that GH has a way to report trademark abuse, or yes of course you can get lawyers involved. Since it's international that might get annoying unless you're fine just getting it off of GitHub and other hosts you can easily reach.
It's also kind of a humorous jab in that it's named "Open".
AI developers don't care about intellectual property.
Which is a positive turn of events. Intellectual property as a concept needs to go.
It's a one-way street with AI companies though, they are against IP protection for training data but support IP protection for the product of that training. Read the TOS/license of nearly any big model and it's clear the creator considers it and its output to be their IP and they will fight to restrict how you can use it for commercial purposes or training competing models, because those being a free-for-all would get in the way of their business model.
In practice it's less "abolishing intellectual property" and more "transmuting everyone else's IP into their own IP so they can be the ultimate IP landlords".
Sure. And I wouldn't be surprised if e.g. Micr.. OpenAI can manage to brib.. convince a judge to make government to enforce this by force.
But how ML is at least currently developing, other megacorporations and universities are only a year or so behind OpenAI etc, and publishing their weights and models. Of course disregarding training set IP because otherwise keeping up would be impossible.
If the media and advertising megacorporations manage to get a ban on training on IP material, Micr.. OpenAI would be a huge benefactor, as they could afford some weird licensing deal on everybody's data and the actual open efforts would be extinguished.
Luckily US hegemony is fading and many other powers DGAF anymore what a US judge says. Or a WTO judge, because USA kind of destroyed WTO themselves already.
If they ban training on IP, it's not unlikely that GPT 4 would be subject to algorithmic disgorgement
Unless the entity they're taking from can defend themselves in court. Just ask Copilot to generate an image in the style of AI hater Tim Burton and you'll get a ToS wrong-think violation. So really it is just for taking from small fries.
Intellectual property in the form of trademarks is generally good for consumers, you don't want a new database product to be called Postgres despite no association with the developers of PostgreSQL.
Intellectual property in the form of patents and copyrights is often harmful for consumers. In music, academia and mechanical engineering it's clearly been harmful in slowing the spread of innovation. It sucks to wait 20 years for a new excellent new dishwasher tray shape.
If you think for more than a few seconds on the matter, you quickly realise that intellectual property is a moral necessity, because intellectual expropriation (which is what you are really talking about) would destroy a major incentive for positive contributions to society.
Everything political is rooted in morality, and while it is easy to point out political challenges on the topic, intellectual property is ethically an unambiguously good concept.
Most of the great art was created in a world without intellectual property, or with no effective IP enforcement on that art.
While I don’t think all IP should be abolished, I think in general there is too much of it, protected for too long, and society would be better off getting rid of patents and dramatically downsizing copyright to at most a few decades of protection.
(And yes I have thought a lot on this matter. The evidence behind patents being overall beneficial is weak, and there is no effective societal argument for the current copyright terms.)
Most of the great art was created in a world without intellectual property
It was also created at a time when copying work was almost as hard and expensive as creating it.
I tend to agree with your takeaway, but I want to pick a nit on your point about great art being made without copyright.
"The Most Powerful Idea in the World" basically makes the case that intellectual property is what enabled the Industrial Revolution. I won't try to summarize or defend the entire thing here, but I think it makes a very compelling case that the notion of "owning ideas as property" was the thing that made Britain unique (among other factors of course) and led to runaway technological explosion. It points out how in earlier times, inventors were literally killed for coming up with better methods that threatened some established system. So there is a big difference between art and technology there, and while I think debating the merits of copyright as it pertains to art is valid (and 100% agree it's overdone in our current system), I'm not convinced the current issues are serious enough to undermine the entire concept of intellectual property.
Enslaving half of the world also played a slight part in that uniqueness.
Most creators work for hire and the company owns the IP. Anecdote: I can't remember the song written on a bus and sold to the record company for $250 to keep food on the table that has earnt billions for Warner, Sony, or Universal Music. Nothing more for the creator. Big company stealing from another big company to benefit the rest of us? Seems like fair use. Nothing to do with creators. That incentive was meant to be something like 20 years (same as patents) not almost "forever" like the Mouse wants who expropriated from all of those European folktales and sanitised them. No incentive if you live off the earnings of a "one-hit wonder".
> sold to the record company for $250
And without any form of copyright it would have gone "hey $RECORD_COMPANY here's this cool jingle I made" and they'd say "yeah we think so too, we'll use that and compensate you $0."
Way better.
A lot of creators, at least in music, are not even work for hire, but in a form of indentured servitude.
I have thought on the matter for millions of seconds.
Politics and economics don't have anything to do with morality. It's about power. The moral fairytailes are there to justify whatever the state of affairs are. When these change, new fairytales justifying them will be adopted.
If right to property is a God given right, why the hell I as a atheist should give a damn?
Examine your definitions of politics and morality. Morality encompasses all actions made by actors with (presumed) free will.
"God given" means "natural". [Natural rights does not require a god](https://en.wikipedia.org/wiki/Natural_rights_and_legal_right...)
And if your argument ("... god given ... atheist ... etc.) can be copy-pasted into a valid defence of arbitrary evil (I like to use Hitler for a concrete example, but avoid it in public because some subset of people erroneously thinks that is an automatic fallacy), then it is trivial in the sense that it doesn't add anything to the discussion, regardless of your opinion or pov.
"Natural" rights are clearly a social construction, and indeed largely the same thing that was theologically justified with God.
They were literally just invented few hundred years ago. And have even changed since then (owning another human being was somehow a natural/god given right). Humans have organized (and keep on organizing) without any such conception for at least tens of thousands of years, but suddenly its natural?
Who is the absolute arbiter of "evil"? USA of course holds itself as a force for good, and e.g. islamism as evil. Islamists do the opposite. Very rarely anybody thinks themselves as evil, but readily point out evil in others.
This is in no way to support islamism, or USA. Just that morality is just mostly an arbtrary justification for whatever somebody wants to or doesn't want to get done.
We can discuss the merits and shortcomings of e.g. ownership based on its societal consequences just fine without conjuring metaphysics.
> If right to property is a God given right, why the hell I as a atheist should give a damn?
You can argue it from political necessity.
Some kind of property rights are necessary for the operation of any social group. Even in a communist society you can't have the air force stealing land from the navy, or the transport department taking electricity from hospitals, or neighbours taking each others food allocations.
There needs to be some kind of organised transfer of property or you get chaos.
You can argue divine right of kings or white supremacy from political necessity.
Some kind of systems to allocate scarce resources is needed. This doesn't have to be property. Access to e.g. the moon and earth's orbits and international waters and polar regions have systems of allocation, but they are not property.
> You can argue divine right of kings or white supremacy from political necessity.
Those arguments can be incorrect without entailing all arguments from political necessity are false.
I can incorrectly argue a justification of the flat earth theory from physics - but that would discredit my own intellect rather than the subject of physics.
> Some kind of systems to allocate scarce resources is needed. This doesn't have to be property.
I agree not everything has to be property, but some things must. Essentially anything that's both important and "stealable" needs to be protected by property rights of some kind.
Regardless of how your food and water come into your possession (trade, charity, gov allocation), it's necessary that some of it remains in your possession for you to consume.
> Those arguments can be incorrect without entailing all arguments from political necessity are false.
> I can incorrectly argue a justification of the flat earth theory from physics - but that would discredit my own intellect rather than the subject of physics.
Do you mean the right to (certain very specific form of) property is somehow an empirically shown fact like the geometry of the earth?
Am I right in guessing that you have inclination towards praxeology?
> I agree not everything has to be property, but some things must.
Why do some things must? Because it leads to more desirable consequences than other options, or because it is some metaphysical truth you're somehow privy to?
The first kind of argument is fine, the latter is just blunt rhetorics. I don't see why the former should be spoiled with the latter.
> Do you mean the right to (certain very specific form of) property is somehow an empirically shown fact like the geometry of the earth?
> Am I right in guessing that you have inclination towards praxeology?
Yes for both. I’ve not formalised it but I suspect we can a draw a line from philosophy -> game theory -> evolutionary dynamics -> socially stable systems of “rights”
I'm all for socially stable systems of "rights". And deriving theories from assumptions is all good. Where I depart from praxeology is that I think the assumptions must be changed when they lead to empirically invalid conclusions.
Also, I don't believe ownership rights like we have currently lead to stable societies. This is kind of core of Marx's critique of capitalism.
I actually think we do live in very socially stable societies. Not to say it’s always a good thing.
But we do have clear systems of who gets power, through what means, and what they can do with it. The various classes of society have a relatively peaceful system of competing with one another.
I’m writing from a British perspective but it largely applies to most developed countries.
Europe had two world wars in the last century, and there's a major war right now. The (private) economy has collapsed at least twice. The post-war stability was largely due to reigning in private propery rights.
From a Finnish perspective, where ownership rights are somewhat lesser, the British system's trajectory didn't seem particularly stable when I lived there 2019-2020, and looks even less so now.
I think too much stability is self defeating. It creates fragility like a monocultured ecosystem. Self correction mechanisms are important, even if they're painful. Bad governments and bad enterprises need to be wiped out now and again by war and recession, as a last resort when other forms of reform fail.
Britain never seems stable at first glance. Read a newspaper from 5, 50, 150 or 250 years ago and they'll all be lamenting economic uncertainty, geopolitical dilemmas and social injustice. Yet in that time no revolutions have killed more than 1% of the population, long term economic growth has averaged couple percentage points annually and nobody has successfully invaded the Islands.
> Some kind of systems to allocate scarce resources is needed. This doesn't have to be property.
I should not be coerced to give output of my labor (whether physical or mental) to others based on some burocrat’s allocation system or my neighbor’s wishes. That would be immoral.
Per your morality perhaps. But morality is just an opinion, and one that can't be justified otherwise.
I should not be coerced not to enter a piece of land just because somebody made a contract with somebody who killed or evicted by force the people (and other animals) who previously inhabited the land. That would be immoral, right?
Why is that the case? The way its implemented is no longer fit for today's world, but as a concept? That's a whole other story.
IP is a huge drag on the progress of mankind.
Ideally there would other ways than economical to facilitate production of intellectual goods. Ownership economies are exceptionally bad in non-scarce resources.
There are some alternative and less destructive ways to handle the economy already in place. Academic institutions generally don't claim or enforce IP (paywalls are more a corporate than academic thing). Many library systems pay authors (notably not the publishers) for each loan from public funds. Open source software is funded via services (and increasingly with ads, which is not necessarily great).
State enforced monopoly on ideas or lumps of matter isn't the only way of structuring the economy. Alternatives haven't been discussed in 30 or so years, so it's understandable that many have hard time even conceptualizing such.
Is IP on names also a huge drag? Would it be beneficial to the mankind if I—or anyone—were able to just make their own Linux or Nginx, with no relation to the group that originally made it?
Arguably "first come, first serve" policy on names per domain is a good one and serves to reduce confusion.
Monopoly on usage of a name not so much, although how it is implemented (or enforced) has substantial problems too. Companies regularly hijack commonly used words as "property" and/or push their monopolized names to the general vocabulary (e.g. iPhone, iPad, App Store, Googling, Kleenex, Xerox) especially in the USA and then literally police how people can use the words.
Yes, they should lose the "property" if they do so, but in practice they lose it decades too late if even then.
Also trademarks being sellable means that at any moment the product with a trademark can have nothing to do with the original holder, partly defeating the rationale.
I think intellectual property is necessary given the current way the economy works. But we should strive for evolving how the economy works.
I agree but in the mean time, stealing from millions of artists and writers to build systems with the goal of replacing them is incredibly shitty.
If we lived in an economic system where people don't starve or lose their house when they don't have a job then I wouldn't care. Until then, OpenAI and its ilk are looters.
> an economic system where people don't starve or lose their house when they don't have a job
I think the AI industry is motivating this for themselves with the idea that this does (at least partially) lead us to that world.
That would give big companies a big advantage.
It maybe wrong how it is now, but it exists for a reason.
Same with patents.
Disney must be close to divine managing to get such a marketshare despite IP regulations holding them back.
Without IP it wouldn't be Disney but the first company with enough power and money would crush any competition.
Even now they can fight in court until smaller opponents go bankrupt but without any IP they wouldn't even have to fight.
Remember when Amazon copied successful products on their marketplace? That would be the norm and you could even blame or sue them.
No. AI firms are a good example of precisely why copyright was put in place.
Right now, AI is taking people's original works and rehashing them in a way that directly competes with the original work. Some AI firms (e.g. perplexity) just have their LLMs paraphrase the work lightly.
This is a problem because it drives original work out of business. Even setting aside matters of originality, artistic value, and AI being vapid slop:
Gen-AI is and will remain a derivative work that is reliant on original human-made work
ChatGPT is not going to do investigative journalism. If we let AI push all journalists into bankruptcy, the news just becomes an endless sewer of PR statements recycled into AI slop.
If you don't want Google's woke "diverse nazis" to push real news into bankruptcy, copyright is needed to stop them.
All (99%) the music I listen to is remixes, all (99%) the writing I read is fanfic.
> Right now, AI is taking people's original works and rehashing them in a way that directly competes with the original work.
Good. Taking people's original works and rehashing them in direct competition fosters, not harms, cultural flourishing.
We should find new ways to fund essential services like journalism and the arts, rather than artificially crimping the possibilities of new technology in order to get a half-assed solution under the current paradigm.
Until we find the political will to do that, creating products that fuck over journalism is shitty.
Interesting that most engineers have this view despite their paycheck largely depending on ip
one of the phrases they used @ their example page maybe reads pretty clear :D
> there are no text or logos in the video
this is a issue
"pretty good" is definitely one point of view... all those videos look very blobby and morph-y
If you want to make an adjective out of an English noun, just add a 'y' directly to the end of it, e.g. 'morphy'.
The hyphen is erroneous here: 'morph-y' would be pronounced like 'morph why', and has the structure of a compound modifier without making any sense as one.
I'm a native English (Canadian) speaker and, given this context, would never pronounce "morph-y" as "morph why" (if that pronunciation was intended I'd capitalize the Y and probably not hyphenate it, "Morph Y").
Indeed, sometimes I'll use the -y construction when I'm inventing an adjective from a noun like this, especially if omitting it looks "wrong" to me because it's an invented word (not sure if I would've for "morphy" but whatever).
I don't know if you're a native speaker and there's some kind of cultural or generational gap here where some of us are more/less particular about this than others or if you're a non-native speaker trying to be helpful but underestimating the flexibility of casual written English.
> I'm a native English (Canadian) speaker and, given this context, would never pronounce "morph-y" as "morph why" (if that pronunciation was intended I'd capitalize the Y and probably not hyphenate it, "Morph Y").
Words conjoined with hyphens are always pronounced separately in normal usage -- the only exception I can think of is the special use case of hyphens representing continuations of a single word across a line break.
> Indeed, sometimes I'll use the -y construction when I'm inventing an adjective from a noun like this, especially if omitting it looks "wrong" to me because it's an invented word (not sure if I would've for "morphy" but whatever).
It's an "invented word" with or without the hyphen. But the standard way of "inventing" word forms in English is to apply suffixes directly to the root word without hyphens.
Using an interstitial hyphen as metadata to indicate one's own squeamishness about using a novel word form seems very strange to me. (The typical convention for this is to enclose the novel word in quotation marks, calling attention to the word itself in addition to invoking its meaning.)
> I don't know if you're a native speaker and there's some kind of cultural or generational gap here where some of us are more/less particular about this than others or if you're a non-native speaker trying to be helpful but underestimating the flexibility of casual written English.
I am very much a native speaker, and it's precisely for that reason that I find this nonstandard use of hyphenation jarring and disruptive of reading comprehension.
So, this morning I happened to be reading a crossword blog - written by an English professor - and, reflecting on this conversation, thought, "Jeez, I'll bet Rex Parker uses the -y construct." So I scrolled back a few pages in his archives, and sure enough, earlier this month he described something as "bro-y" [0].
In case you still feel like you're more of an authority on whether this is acceptable in colloquial English than me (a Canadian native Engish speaker), OP (who - I peeked at their profile - appears to be a British native English speaker), or an American professor of English:
I also noticed, while scrolling through Rex's archives, that the Times itself had used the "-y" in the crossword. See the cited clue on 36A on June 18 [1], which describes something as "sting-y" - not to be confused with "stingy". Good enough for The Gray Lady, good enough for me.
At this point, I think - if it's still causing you to struggle with reading comprehension - you need to accept that it's a construct in common usage and spend some time practicing until you start to feel comfortable with it. Sorry, I know it's inconsistent; English is known for having quite a few quirks that you just need to get used to.
[0] https://rexwordpuzzle.blogspot.com/2024/06/strong-as-ox-in-s...
[1] https://rexwordpuzzle.blogspot.com/2024/06/extremely-damaged...
As I mentioned in another comment, I think using the hyphen to eliminate ambiguity where it would otherwise create a word form that's easily confused with, or sounds like, another word entirely is perfectly reasonable. The "sting-y" and "bro-y" examples seem consistent with this, but "morph-y" doesn't quite.
That doesn't always work in a way that's clear, though. If I said a microcontroller felt pretty army, that's different to saying it feels arm-y, since army is already a word. I wasn't trying to say that it was like chess player Paul Morphy, I was trying to indicate it was an adjective derived from "morph". Language is meant to convey meaning, and you understood the meaning I was trying to convey, so what's the issue?
> If I said a microcontroller felt pretty army, that's different to saying it feels arm-y, since army is already a word.
I'm assuming that you don't mean that the microcontroller reminds you of an upper body appendage, and are instead referring to a well-known instruction set architecture. If that's the case, I'd recommend 'ARMy'. But you're correct about the ambiguity where the word form you want to use collides with another pre-existing word. In that situation, you'll naturally need a different solution.
> Language is meant to convey meaning, and you understood the meaning I was trying to convey, so what's the issue?
That's precisely it. Language is meant to convey meaning: and the more complex and abstract the meaning we mean to convey, the more precise we need to be in our use of language. Deviating too much from well-established conventions makes ambiguity or miscommunication more likely. Conversely, to your point, blindly adhering to conventions where that would increase ambiguity also impedes communication.
In this case, the erroneous hyphen didn't completely erase your intended meeting, but it did make the sentence slower to parse.
Actually you're meant to write "morphey". But the original is clear enough.
I was forwarded a message just moments ago by my mother, a 5 second Facebook video, clearly AI generated - and I suspect Open Sora.
More AI slop for everyone.
At some point we should stop treating our mothers as infants. It doesnt matter that they didn't grow up with the internet, they should adapt , despite their age. It has happened before and will happen again.
(it makes me mad that similar arguments are often brought when discussing mobile phones' walled garden policies)
Oh I told her what it was, she said she'd put her reality glasses on. We'll see how it goes :)
old people don't struggle with tech because they didn't grow up with it
well, they kinda do, but here's the thing
when you're old, all the tech that's out will be alien to you
I'm not even middle aged, I'm a tech enthusiast, and nonetheless it is happening to me already. new UI and new tools all happening too fast to keep up
we should be trying to do something about this for ourselves, not just for our mothers
Sure, but if it s dangerous to you, then maybe you shouldn't be using it - at all. We trust older people to vote, they should do it in good judgement and this includes consideration of where they get their information
I know AI will be used to spam, but to not see or acknowledge the positive capabilities of this technology is absolutely flabbergasting to me.
This is going to make filmmaking cheap and accessible to a broad audience, and that's nothing short of a miracle.
I typically do each year's 48 hour film project with a team of 8 or more people. This year I did it with myself and an editor, and it was amazing - no 6 AM trip to the prop house, no lifting heavy props from the top shelf, no signing location release forms, no sweltering lights. I didn't have to put on makeup or do blocking for hours. (I used mocap, rotoscoping, and compositing.)
I've been using AI diffusion canvases to design - something I'd previously found entirely intimidating and unapproachable. I'm having so much fun and enjoyment with this.
I don't see the world the same way, and I don't get this negativity.
I get it. I play with image generation myself, with the odd bit of training and fine-tuning. But you have to recognize that the low marginal cost of this kind of generation is self-cheapening, and a world filled with gaudy saccharine content (the kind of stuff that wins RLHF head to head comparisons) upvoted and shared by people who don't recognize it for what it is, is a little bit dispiriting.
You've hit the nail on the head. Its good at designing. But you can see from the visual quality of the examples that its not ready to be used as final content. No AI content I've seen is really ready to be used as a final piece, It shouldn't only ever really be the basis for something. I have no issues with an AI generated piece being used as a "sketch" or a "storyboard" but using it for a final product is just... lazy and looks crap?
Can we see your this years project anywhere? Just curious to what you achieved with it, how complex the film is etc.
How can they use the Sora name, isn’t that trademarked?
You can do anything you want until you get caught doing it.
Quite possibly: https://bike.shimano.com/en-US/product/component/sora-r3000....
The domain matters. You probably can sell a car called Sora, but not AI video models or bike parts.
Just a name
It actually seems to look pretty terrible.
IMO it's "pretty good" from a technical and power efficiency perspective, but to someone who doesn't know/understand/care or has to deliver a "product" this is far from good or acceptable.
Having said that, it's pretty impressive.
One thing I wish for is for AI-generated images to have awareness for physical constraints.
Genuine question: can you just call your alternative to product x "Open-x"?
Ironically, I think having a Twitter clone be called Open-X might fly legally, since that's generic enough, but pretty much any other instance where the trademarked name is very specific, like Open-Google or Open-Facebook would be surely easy to be removed by a judge. It's just that it takes a while for this process to happen, and until there's a legal injunction you can still use the name.
I bet that this will result in a letter from a lawyer, if this gets any traction.
Depends on how rich and powerful you are.
Usually, no. If they have a registered trademark, you definitely can't.
Assume "pretty good" here means "pretty good relative to Sora", because even Sora can't claim to be "pretty good" yet.
You could say that actually being available to the public makes it pretty good relative to a product that's still vaporware 5 months after being announced. There's no timeline for Sora availability, or any indication of its compute demands, and by extension its cost per generation.
I don't think vaporware as a term really applies here. Sora is real and I've seen it used live - it actually works and it is pretty astounding. I don't know how well scaling it for the public will go but they definitely built the thing.
Why do you think so? The videos looked very impressive for the most part IMO.
github repo: https://github.com/hpcaitech/Open-Sora
Anybody else completely apathetic towards all of this? I'm so fed up with the morally-bankrupt Sam Altman and Open AI as a whole that by the time another one of these "Hey, check out the latest blah blah AI blah!!" gets posted, my first (and often lasting) impression "meh".
For me it's more than apathetic - for example when I'm encountering content which is illustrated with AI-generated images I noticed that I'm subconsciously downgrading it's quality right away - for sure it's wrong because that's classic example of judging book by the cover and even some good quality OS software on GH and/or good tech blogs with proven record use AI-generated content for illustrative purposes - however it's kind of automatic reaction for me and I need to use conscious effort to combat it.
I think it’s reasonable. You weigh the quality of content by the amount of effort its creator put into creating it.
Has nothing to do with AI directly.
The dramatic predictions that tend to occur between the model updates are what annoy me. AGI is coming! It’ll be smarter than PhDs! We will need technocommunism!! Cold War 2!
These look like shit. Yeah I guess it's kinda cool you can make them with nothing but text but I dunno man, I could think of a dozen ways to make all of these shots that wouldn't have the sides of every object threatening to melt in the next frame. More difficult and cost more, I mean yeah, but again, then it won't look like a computer is hallucinating your scenes while also tripping on mushrooms.
If this is indeed the "future of making movies," I dunno, I can't speak for anyone but myself, but I have no interest in watching anything made this way. Sorry. If you don't care enough about your project to do something besides AI prompts, I don't think you have anything to say that I need to hear.
> If this is indeed the "future of making movies," I dunno, I can't speak for anyone but myself, but I have no interest in watching anything made this way.
I'm still burned out on "regular" CG in movies. I'd be much more excited about a (hypothetical) return to practical effects than a push even further into computer imagery.
I recently watched John Wick 1 and I was mostly stricken by how little CGI was used in the movie.
I kept watching, noticing small bits here and there, subtle things, but there was not much CGI compared to almost any other action movie I've seen in the last several years, especially not much of any obvious full green screen sets and fly-by-wire wuxia heroism BS.
Anybody else a bit scared for how this tech will impact people? It seems quite far reaching. Job loss from video production and film. Ease of spreading misinformation, lies and propaganda. I could even see less investment in physical infrastructure as real world experiences get replaced by generated video. Scary stuff.
A hammer can be used to hit nails or bash skulls. There are billions of hammers.
This is more like a magic wand in terms of its effectiveness . I think that’s the difference.
Since we were talking about hammers, a more apt comparison might be a nail gun. Vastly more efficient than hammering nails by hand when constructing e.g. a roof truss, but you still need carpenters.
Hammers don't kill people, people kill people
... with hammers
If you want to look at it from a brighter side, I think it's great that this tech exists and people know about it in large and how easy it is to fabricate false information with this.
If I was very optimistic, I'd hope that people would question more what they see and hear and read to due to widespread of "Gen AI tools".
In reality that will most likely not happen, but yeah, one can hope.
And on job loss: Jobs come and go, nobody cared much about carriage drivers when cars took off...
Exactly this. What if we can now produce 100x video content and anyone can do it? YouTube didn’t kill TV shows just because anyone could make a video or stream live.
Yeah but the people reskilled into other demanding jobs that still needed people.
With AI, the demand for that skillset is diminishing as is the need for people at all.
We're heading towards a Solaria situation, where the rich genuinely might no longer need to employ the masses, or sell them products. What would you do if you had a magic box that met all your needs? Pay a consultant? Hire a gardener? Call a maid? Request an escort? Why.
AI is not and most likely will not be a "magic box that meets all your needs" in at least 1000 years unless something insane happens, AI does function approximation, when you scale it up it can do amazing things but you still need human creativity, you still need engineers because it can only output data not act on real-world stuff, you still need data that is generated by humans, you still need to train every model for your specific task, and tasks are infinite.
If we wanted to replace humans we'd have to fully figure out how the human brain works first, for now no one knows how the human brain truly works so how can we claim AI will replace all humans?
In my opinion, AGI is just an acronym pushed by people who have something to gain from mass hysteria, FOMO, and making people feel like "the end is near", AKA, AI influencers and founders who want investors to give them money.
Of course, this doesn't mean everyone who believes in AGI is one of the two, most were probably just fed this narrative and it sounded believable so they just went with it.
These are my two cents, sorry for the wall of text :P, but I just had to say something about what I perceive as an "AI bubble" that generates mass panic for most in order to benefit a few
Many of the day-to-day tasks that keeps people regularly employed are not creative tasks, but repetitive and predictable.
Sure, repetitive tasks may get replaced in the short time depending on the complexity, but, the way I see it, there will always be a human need, new jobs of higher complexity and creativity will be created, and maybe there will be less jobs overall, but predictable jobs being replaced always happen with new advances in technology, whether that's a good thing or not :)
> always be a human need
In what quantity though? In Willy Wonka, when Charlie's dad got laid off at the toothbrush factory and was later rehired as the sole toothbrush automation inspector, what happened to his colleagues? Bigger and better things? Unlikely, unless their kids also suddenly inherited chocolate factories.
It's a decimation of the workforce, and we're still climbing the population curve.
> What would you do if you had a magic box that met all your needs? Pay a consultant? Hire a gardener? Call a maid? Request an escort? Why.
Because that magic box is… magic, fantasy.
Terrified.