• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

OpenAi announces Sora, new text to video AI generation tool

ResurrectedContrarian

Suffers with mild autism
Wait, I don't get it how this works. How are the videos generated? They are CGI but where are all those 3d models and textures coming from?!
Models and textures are not involved. That kind of classic CGI tech is rapidly becoming obsolete.

This is a diffusion model; so it samples novel visuals in a probabilistic way from its internalized visual knowledge of the world based on its massive training data.
 

StreetsofBeige

Gold Member
I thought it would take another 5 year for amazing AI video. This will absolutely break the advertising market + movie industry.

I'm all ready for an our robots AI overlords and become a pudgy doing nothing on the couch.
Yup.

Just like bloggers and every day joes getting popular on the net, taking away from traditional media industry jobs, at some point you'll get every day joes making cool videos. And then, the person will make their own 10 part series, then when the tech is there a 2 hour movie etc...

Does it have to be perfect? Nope. All the normal people on Tik Tok or YT arent either. But as long as it's entertaining, anything can work in media. Just need someone to make it and show people.
 

StreetsofBeige

Gold Member
Cant wait when I can feed it a book and give guidelines what it should look like visully.
All this AI stuff should be a boon for writers IMO. I'm not a book reader. I might read tons of stats like business and sports stats, but for traditional 350 page books no thanks. Too boring and endless walls of text. Books I have at home usually have pictures.

But I can see writers adding lots of AI art or AI clips to their digital book to spruce it up. Maybe every 10 pages is a 60 second clip.
 

Chittagong

Gold Member
They really need to connect the hallucination and human input better to make generative art truly useful. I can get amazing imagery out of Midjourney, but it takes a lot of rerolls and changing ways I ask for things, and just chance. Creating better creative direction tools will unlock the potential of this tech.
 

ADiTAR

ידע זה כוח
Don't worry! If there's one thing that unelected academics and scientists and fact checkers have proven in the last several years, it's that they can be trusted to decide for everyone else what people should be allowed to say and hear and read.

Because elected officials really have our interest at heart...
 

Dr.Morris79

Gold Member


Make It Stop My Brain Hurts GIF by Monty Python

You're not kidding. LSD users are in trouble, that's for sure.
 

Aggelos

Member
Sam said that we are on this continuous upward trend, this steep improvement curve. So it's happening...



 
Last edited:

Dr.D00p

Gold Member
every other job that uses the brain (lawyers, accountants, bankers, scientists, programmers, engineers, doctors) can all be replaced with AI.

If those class of people are threatened like that, they will find a way to legislate their way to protection.

The 1% will always find a way to keep themselves on top.
 

efyu_lemonardo

May I have a cookie?
South Park was right on with their prediction, the last 150 years have been where people who use their brain out earn people who provide physical labor.
But with AI we could be returning to the old days. A handyman who can fix stuff physically cannot be replaced by a computer, every other job that uses the brain (lawyers, accountants, bankers, scientists, programmers, engineers, doctors) can all be replaced with AI.


Physical labor is probably no more than a decade behind intellectual one.
 

Lord Panda

The Sea is Always Right
Cant wait when I can feed it a book and give guidelines what it should look like visully.

Except that it will be censored to shit. Anything towards the risque, conflict, controversial, copyright etc will require a frustratingly amount of effort to circumvent.

For example it took me ages to generate this image of Lara Croft, because DALL-E didn't want to infringe potential copyright, and preferred to do its own thing instead of listening to my directions. I had to trick it into generating this picture (for the Crystal Dynamics next Lara Croft thread):

Lara.png


As another poster stated, it will only be truly be useful once these limits are removed (fuck off 'red team'), and/or run locally without restrictions and under our full control.

One day I'm hoping we will have the tools so that I can fix Peter Jackson's LotR trilogy. The first thing I'd fix are the goddamn army of the dead scenes.

Maybe this is why the Enterprise didn't have video surveillance 🤔?

We need more advanced sensors and scanners or nothing will be admissible in court.

I'm assuming that the meta-optics / metamaterial sensors of the future will provide so much fidelity and metadata that perhaps it will be pretty straightforward to distinguish between real images and AI generated images.
Metamaterials, with their ability to manipulate electromagnetic waves, could lead to sensors that capture far more information than current imaging technologies. This could include not just visible light but other spectra, polarisation information, and even the detection of subtle environmental cues that are invisible to traditional cameras. There's a great video that covers one such metamaterial:



Though it's also likely that AI generated imagery will in time also be able to spoof the relevant data as well. I wouldn't be surprised to see AI agents being used to help determine the authenticity of a given media, like a virus scanner.
 
Last edited:

Thaedolus

Gold Member
This is terrifying dystopian shit, all I see is this is going to be used for all the wrong reasons.
The only hope I have is that some more intelligent AI will be able to detect if something is AI generated or not, but then again, who’s going to trust it and/or the people telling us “no really the box says this is fake so it’s fake!”

When like one in five people already believe Taylor Swift is a psyop…
 

LimanimaPT

Member
Models and textures are not involved. That kind of classic CGI tech is rapidly becoming obsolete.

This is a diffusion model; so it samples novel visuals in a probabilistic way from its internalized visual knowledge of the world based on its massive training data.
Really? Fuck... Now I'm impressed. This can be a major breakthru in video game development. Maybe in a few years games will no longer take years to develop.
 

ResurrectedContrarian

Suffers with mild autism
Really? Fuck... Now I'm impressed. This can be a major breakthru in video game development. Maybe in a few years games will no longer take years to develop.
What's really interesting is that you can use these AI models, and their internalized understanding of the visual world, and get them to create 3D models or fully rotatable 3D views of imagined objects and scenes, and then use these in other tasks like import them into real time engines or a game etc.

For some very simple examples:




But for more mindblowing work, look up neural radiance fields : )
 
Last edited:

efyu_lemonardo

May I have a cookie?
So what then? Society will collapse when 90% of people have no purpose (for society) and no income.
Wish I knew. I wish leading expert economists were urgently attacking this problem around the clock.
Even without AI and robotics, the global population is on course to stop growing by around 2050. That's a single generation from now, and it's something that no modern economy is built to deal with. The entire pension and retirement systems are designed around positive population growth.

So it could be that in the short term - i.e. the next few decades - we'll see AI and robotics fill the labor gap created by lower birth rates, in order to sustain current economic models. But eventually - likely some time later this century - there will have to be a monumental shift in how we treat capital and how we define growth.

My hope, as a teacher, is that in a world with much fewer children, the social value of good education will increase significantly, and the percentage of resources spent on education will grow. Hopefully this will lead to a world with less ignorance and better values, but there's no way of knowing what will happen that far into the future.
 
Last edited:

Bojji

Member
Wish I knew. I wish leading expert economists were urgently attacking this problem around the clock.
Even without AI and robotics, the global population is on course to stop growing by around 2050. That's a single generation from now, and it's something that no modern economy is built to deal with. The entire pension and retirement systems are designed around positive population growth.

So it could be that in the short term - i.e. the next few decades - we'll see AI and robotics fill the labor gap created by lower birth rates, in order to sustain current economic models. But eventually - likely some time later this century - there will have to be a monumental shift in how we treat capital and how we define growth.

My hope, as a teacher, is that in a world with much fewer children, the social value of good education will increase significantly, and the percentage of resources spent on education will grow. Hopefully this will lead to a world with less ignorance and better values, but there's no way of knowing what will happen that far into the future.

I'm sure corporations will see short term gain: "we can replace 3000 workers with robots and make massive savings!", but the problem is these workers are also (potentially) their consumers and with every business thinking like that there will be not much consumers left.

Universal income?

I think we are ahead of financial crisis that will make 2008 crisis look like a joke.
 

efyu_lemonardo

May I have a cookie?
I'm sure corporations will see short term gain: "we can replace 3000 workers with robots and make massive savings!", but the problem is these workers are also (potentially) their consumers and with every business thinking like that there will be not much consumers left.

Universal income?

I think we are ahead of financial crisis that will make 2008 crisis look like a joke.
It's not merely a matter of less income in the hands of consumers, either. People with nothing to keep them busy and with no meaningful social calling will become more prone to radicalization and other forms of anti social behavior.
 

Bojji

Member
It's not merely a matter of less income in the hands of consumers, either. People with nothing to keep them busy and with no meaningful social calling will become more prone to radicalization and other forms of anti social behavior.

Yep, usually people that don't work (at all) are the most problematic in society.

Like you said this should be monitored by economists and governments 24/7 but of course they will wake up when something very bad happens.
 

Pagusas

Elden Member
Really? Fuck... Now I'm impressed. This can be a major breakthru in video game development. Maybe in a few years games will no longer take years to develop.
Yep, im so excited for the future of gaming is going to be, truly responsive dynamic worlds that are altered by your actions, not by pre-defined scripts or programmers intentions, but rather worlds that have fully ai responsive character dialogue, changing landscapes, changing stories and characters. Think Star Trek holodeck episodes without the fancy holodeck part.
 

efyu_lemonardo

May I have a cookie?
Yep, im so excited for the future of gaming is going to be, truly responsive dynamic worlds that are altered by your actions, not by pre-defined scripts or programmers intentions, but rather worlds that have fully ai responsive character dialogue, changing landscapes, changing stories and characters. Think Star Trek holodeck episodes without the fancy holodeck part.
Aren't you concerned with the seemingly increasing gap between the quality of virtual life and the quality of physical life?

Will people increasingly disengage from the real world causing it to slip further into neglect?
 

Pagusas

Elden Member
Aren't you concerned with the seemingly increasing gap between the quality of virtual life and the quality of physical life?

Will people increasingly disengage from the real world causing it to slip further into neglect?
I've not seen that, maybe because we're well off and travel the world twice a year and see the most amazing places every few months, but the world is still amazing and the quality of everywhere we go is still great. Same with where we live. I'd be curious to see what you are seeing, any examples?
 

efyu_lemonardo

May I have a cookie?
I've not seen that, maybe because we're well off and travel the world twice a year and see the most amazing places every few months, but the world is still amazing and the quality of everywhere we go is still great. Same with where we live. I'd be curious to see what you are seeing, any examples?
Housing is more expensive, food and gas are more expensive, and with these kinds of technologies many jobs will become less secure. Quality of life will go down for many, I don't see any way to avoid it.
 

hyperbertha

Member
All this AI stuff should be a boon for writers IMO. I'm not a book reader. I might read tons of stats like business and sports stats, but for traditional 350 page books no thanks. Too boring and endless walls of text. Books I have at home usually have pictures.

But I can see writers adding lots of AI art or AI clips to their digital book to spruce it up. Maybe every 10 pages is a 60 second clip.
Um no. I prefer books without pics thank you. Your potato attention span is your problem alone.
 

Dutchy

Member
Cool stuff but until we can alter the raw 'data' it's going to be pretty useless unless you want a video consisting of shots just barely long enough for the average person to not spot any goofy mistakes.
 

Jinzo Prime

Member
I'm assuming that the meta-optics / metamaterial sensors of the future will provide so much fidelity and metadata that perhaps it will be pretty straightforward to distinguish between real images and AI generated images.
Yep, we will need more data to distinguish between what is real and what is generated. I'm trying to imagine what a "realistic" murder trial would look like in Star Trek, for reference.
 
Read some interesting analysis that suggests this model might be heavily trained on synthetic data - meaning for instance that they model a person walking down a street in UE5, then record it from a ton of different angles and add comprehensive metadata to it. Don't know if it is true but would make a hell of a lot of sense.
 


Physical labor is probably no more than a decade behind intellectual one.

Physical labor will always be behind the intellectual one because it will never scale as well.
You can add the equivalent of a million additional process being run to replace a million programmers by just adding some ram and some microchips.
Whereas even adding ten thousand robots require much more work because you first have to mine the various ores from the ground, extract the raw metal, manufacture them into parts/wiring, assemble it in a factory, and maintain and repair them as the wear down, as well as maintain and repair the factories, the mining operations, etc, etc. The limitations of our physical reality slows down the rate of expanding capacity.
 
Last edited:

Portugeezer

Member
Awesome tech. One day AI will be making a complete video game: "make Hollow Knight 2 silksong"

Scarily fast for anyone involved in the arts.
 
Impressive as it looks, it does have limited applications in production because if you tell it to render a different clip with the same character, it will not be exactly the same, or rather, there are enough subtle differences that will create continuity issues. It's fine for making one-off clips but the pipeline doesn't work like that films/tv.

The other demo I saw was a clip of woman talking down a streets of Tokyo. That video highlights the things it has difficulty with, such as text, and animation. The motion of the camera and the woman is very stiff and robotic like. It's uncanny valley.
 
Impressive as it looks, it does have limited applications in production because if you tell it to render a different clip with the same character, it will not be exactly the same, or rather, there are enough subtle differences that will create continuity issues. It's fine for making one-off clips but the pipeline doesn't work like that films/tv.

The other demo I saw was a clip of woman talking down a streets of Tokyo. That video highlights the things it has difficulty with, such as text, and animation. The motion of the camera and the woman is very stiff and robotic like. It's uncanny valley.
Of course, but this is just the baby stages. Remember we went from 8 bit pixel art to Toy Story in less than 20 years.
 

Audiophile

Gold Member
LINK (embed won't work)

It's the mundane stuff like this that's blowing me away. The temporal stability mixed with the near-perfect interaction between the cat, the person and the bedding.

Other than an extra paw appearing and her brow being a bit squishy, this is just absurdly good.

The fact that we've gone from low res, mushy trash stills in dall-e to this in two years is mind-boggling.

I wouldn't be surprised to see games hybridising specialised versions of this with traditional engines.
 
Last edited:

Audiophile

Gold Member
Porn is going to get real freaky.
Given that they can already animate 3D models of people from still photos; when combined with this and more progress, people are gonna be railing their favourite celebs in lifelike 3D environments in VR in a matter of years.

This stuff's gonna get really squirrely...!

This isn't just disruptive to creatives and industry, to fact and fiction, but to fundamental human decency, privacy and agency over your entire being.

It's happening and there's no getting around it; and soon enough this power will trickle down to the hands of individuals. The mere act of people having themselves in the public eye in any respect will be of great consequence.
 
I think pretty soon we will enter a society where everything online is essentially the virtual world. You don't assume anything you see on a screen as reality, but a alternative dimension. Like a comic book or a fictional tv show.
The only thing people will accept as real are things the can actually see in person.
 

IDKFA

I am Become Bilbo Baggins
Sheit man... how long until I can upload an image of myself with a voice sample and then have Mission Impossible play with me as main character instead of Tom Cruise?

This is wicked.

Honestly? Probably 2-3 years. Maybe sooner.

It's advancing at such a rapid pace it's really impossible to say.

In 5 years we could be designing our own video games from start to finish with a series of text prompts. Scary times ahead.
 
Top Bottom