OpenAI unveils "Sora." A prompt-based short video generator with amazing results

[JustLovett0]

 - - Summary

OpenAI announced “Sora,” a text-to-video generator that will take prompts in a similar way to Dall-E to create short video clips.

 

Quotes

Quote

Sora is able to generate complex scenes with multiple characters, specific types of motion, and accurate details of the subject and background. The model understands not only what the user has asked for in the prompt, but also how those things exist in the physical world.

 

 - - My thoughts

If these showcased prompts are what you would typically expect to see out of the Sora video model, OpenAi is VERY quickly advancing the quality of their machine learning’s results. It is still in the early testing stages and not available for public testing in any capacity, but I am VERY excited to see the limits of what Sora can generate. Good to see the typical “ai look” of videos morphing constantly might just be fixed.

 

Sources

 https://openai.com/sora#research 

[StDragon]

Can't wait to animate this. All this AI technology just for the lulz.

[tim0901]

This is fucking terrifying.

[DuckDodgers]

Well, there's the future of computer graphics Nvidia was probably dreaming all the time. Low resolution input model -> nextgen DLSS inference -> 4K+ 120Hz output.

[StDragon]

18 minutes ago, tim0901 said:

This is fucking terrifying.

Only for Hollywood.

Finally we'll have good content to watch again.

 with extra  please.

I for one welcome our information apocalyptic overlords.

[creat0r]

So much AI generated videos of political figures saying things they never said will surface when these models can generate audio. Humanity will just universally have to agree to never trust anything shown on video anymore. Imagine if Russia made an AI video of Biden saying he'll begin bombing Russia in 5 minutes to fuel the Russian people to protect themselves from a threat that doesn't exist, and therefore start a war. Police will no longer be able to use video tapes as proof of anything, since it might just be AI generated.

[Ydfhlx]

1 hour ago, DuckDodgers said:

Well, there's the future of computer graphics Nvidia was probably dreaming all the time. Low resolution input model -> nextgen DLSS inference -> 4K+ 120Hz output.

Good luck running that model with 8GB of VRAM and consumer chips. Not to mention latency issues.

 

Nvidia was dreaming about it because margins they make on cards actually capable of training and running these models are insane.

[Uttamattamakin]

The obvious implications for all content creators really might be an issue.  There will be deep fakes of adult content and people doing it who never did.    For that matter.... remember the Endless Seinfeld forever "show".  There could be a version of that that looks as good as an episode of Curb Your Enthusiasm.    Endless Curb Your Enthusiasm. 

[Kisai]

7 hours ago, JustLovett0 said:

 - - Summary

OpenAI announced “Sora,” a text-to-video generator that will take prompts in a similar way to Dall-E to create short video clips.

 

Quotes

 

 - - My thoughts

If these showcased prompts are what you would typically expect to see out of the Sora video model, OpenAi is VERY quickly advancing the quality of their machine learning’s results. It is still in the early testing stages and not available for public testing in any capacity, but I am VERY excited to see the limits of what Sora can generate. Good to see the typical “ai look” of videos morphing constantly might just be fixed.

 

Sources

 https://openai.com/sora#research 

Cherry picked.

 

Please... please stop falling for this. For every passable thing generative AI puts out, there are thousands of rubbish results, and the reason companies are so invested in this is because people have to pay for the garbage it generates too. So it's seen as potentially very profitable if it works, so they will show all the cherry picked stuff and not tell how many mutative steps were required to get that result.

[Lightwreather]

It appears Larry the Cucumber was right

[StDragon]

9 hours ago, Kisai said:

For every passable thing generative AI puts out, there are thousands of rubbish results..

So much this! Generating content with AI is basically a roll of the 

You know what's truly comical? Meme generating content with 90%+ having misspellings and bad punctuations. 

I should have included those; such a missed opportunity.

[Brooksie359]

16 hours ago, creat0r said:

So much AI generated videos of political figures saying things they never said will surface when these models can generate audio. Humanity will just universally have to agree to never trust anything shown on video anymore. Imagine if Russia made an AI video of Biden saying he'll begin bombing Russia in 5 minutes to fuel the Russian people to protect themselves from a threat that doesn't exist, and therefore start a war. Police will no longer be able to use video tapes as proof of anything, since it might just be AI generated.

I could be wrong but I have heard that one of the easiest way to figure out if something is AI generated is by using another AI to analyze it. Not an expert in AI so not sure how true that is but I wouldn't be surprised if that was the case especially because I know we already use AI to detect bots vs real humans in alot of cases. Once necessary I think we will probably have some AI tool that will be able to verify if a video is AI generated. I can think of alot of goverments and companies that would want such technology so the chances of such a tool being developed is very high imo. 

[Shreyas1]

23 hours ago, StDragon said:

Only for Hollywood.

Finally we'll have good content to watch again.

 with extra  please.

I for one welcome our information apocalyptic overlords.

Remember that generating fake video is only half of it. If I set up a security camera and catch a suspect that was stealing my stuff, they can just say in court that the evidence was ai generated. 
 

could this technology make security cameras obsolete? That would definitely be scary for some. 

7 hours ago, Brooksie359 said:

I could be wrong but I have heard that one of the easiest way to figure out if something is AI generated is by using another AI to analyze it. Not an expert in AI so not sure how true that is but I wouldn't be surprised if that was the case especially because I know we already use AI to detect bots vs real humans in alot of cases. Once necessary I think we will probably have some AI tool that will be able to verify if a video is AI generated. I can think of alot of goverments and companies that would want such technology so the chances of such a tool being developed is very high imo. 

At least based off ai plagiarism detectors, there's also a big false positive rate to take in mind. 

[Brooksie359]

23 minutes ago, Shreyas1 said:

Remember that generating fake video is only half of it. If I set up a security camera and catch a suspect that was stealing my stuff, they can just say in court that the evidence was ai generated. 
 

could this technology make security cameras obsolete? That would definitely be scary for some. 

At least based off ai plagiarism detectors, there's also a big false positive rate to take in mind. 

Not sure what AI plagiarism detector is but I would imagine the more info you have the better it would be at detecting and video has a ton of information that could give clues that it is done by AI. I mean the checkmark that you aren't a robot is really good at determining if someone is a robot or not with very little false positives as far as I am aware. It only uses how your mouse moves to the checkmark to determine if it was done by an actual person or a robot. I believe they also used something similar for detecting sophisticated aim bots that are supposed to mimic real people but it still is easily detected when using AI. One key component they did say for the aimbot was that while some people might do things that might indicate they are using an aimbot when they aren't that is only a single data point and the chances of a person consistently looking like an aimbot is statically very very low. That is why I would imagine with AI video it would be the case that you would have less false positives because you would have to have it flag throughout the video and not just a small portion. 

[Brian McKee]

1 hour ago, Shreyas1 said:

 

could this technology make security cameras obsolete? That would definitely be scary for some. 

 

The implication would be that AI image/video output doesn't have identifiers. We're a very longs way off of creating things indistinguishable from reality. Right now you can only fool people who don't really pay attention.

[StDragon]

11 hours ago, Brooksie359 said:

I could be wrong but I have heard that one of the easiest way to figure out if something is AI generated is by using another AI to analyze it. Not an expert in AI so not sure how true that is but I wouldn't be surprised if that was the case especially because I know we already use AI to detect bots vs real humans in alot of cases. Once necessary I think we will probably have some AI tool that will be able to verify if a video is AI generated.

Many solutions have been made, but really it comes down to digitally signing images from source material such as DSLR and phone cameras.

There's no point in forcing AI sources to sign images because it's super easy strip the signing out by washing the image through another open source application and resaving it as another format.

So what that means is that any image that's not signed from authoritative source hardware (no point in signing with software) is automatically suspect and deemed not trustworthy.

[LAwLz]

On 2/16/2024 at 12:58 AM, creat0r said:

So much AI generated videos of political figures saying things they never said will surface when these models can generate audio. Humanity will just universally have to agree to never trust anything shown on video anymore. Imagine if Russia made an AI video of Biden saying he'll begin bombing Russia in 5 minutes to fuel the Russian people to protect themselves from a threat that doesn't exist, and therefore start a war. Police will no longer be able to use video tapes as proof of anything, since it might just be AI generated.

This is not a new threat. If anything, I think it would be good if people became more skeptical of what they saw in videos.

We already have the capability and have had it for a long time, to take clips out of context, dub over them, or entirely fake them to a very convincing degree. 

People already pick and choose which things to believe, and how to interpret them.

 

 

On 2/16/2024 at 5:22 PM, Brooksie359 said:

I could be wrong but I have heard that one of the easiest way to figure out if something is AI generated is by using another AI to analyze it. Not an expert in AI so not sure how true that is but I wouldn't be surprised if that was the case especially because I know we already use AI to detect bots vs real humans in alot of cases. Once necessary I think we will probably have some AI tool that will be able to verify if a video is AI generated. I can think of alot of goverments and companies that would want such technology so the chances of such a tool being developed is very high imo. 

The problem I see with that is that the AI bot might just flag things incorrectly.

A lot of people seem to talk about how they can't trust chatbots to get things correctly, yet when it comes to using chatbots to detect things written by chatbots all the skepticism seems to get thrown out the window.

[Brooksie359]

9 minutes ago, LAwLz said:

This is not a new threat. If anything, I think it would be good if people became more skeptical of what they saw in videos.

We already have the capability and have had it for a long time, to take clips out of context, dub over them, or entirely fake them to a very convincing degree. 

People already pick and choose which things to believe, and how to interpret them.

 

 

The problem I see with that is that the AI bot might just flag things incorrectly.

A lot of people seem to talk about how they can't trust chatbots to get things correctly, yet when it comes to using chatbots to detect things written by chatbots all the skepticism seems to get thrown out the window.

I know they use this for detection of aimbots and it worked very well because they would see if it the players movements were flagged more than once. I would imagine it would be the same here. If you have the ai flag things that are indicative of it being AI generated and it flags only once then it probably is a false positive while if it flags it a bunch of times then it is like AI generated.

[TetraSky]

That's crazy. AI will only ever get better and better, taking jobs away from people.

Who needs fiver for a short video when you can just get an AI to make it even better.

 

It's really unfortunate that no jobs are safe from AI and our robot overlords... Even the oldest job in the world won't be safe for long.

[LAwLz]

4 hours ago, TetraSky said:

That's crazy. AI will only ever get better and better, taking jobs away from people.

Who needs fiver for a short video when you can just get an AI to make it even better.

 

It's really unfortunate that no jobs are safe from AI and our robot overlords... Even the oldest job in the world won't be safe for long.

I see that as a good thing. The end goal of humanity should be that nobody has to work. That everyone has been replaced by robots.

The big issue is that society might not adapt at a fast enough rate, so we will have some unknown number of years with issues.

 

 

Another way you can view this is also that it is democratizing the creative process. People who used to have to rely on others to create things can now freely bring their vision to life. That's great.

I want everyone to be able to create images and video, or be able to understand other languages, just like I want everyone to be able to read and repair their own products.

Being against letting people use AI to generate art for example is in my eyes like saying YouTube should remove all home renovation tutorials because it reduces the demand for contractors and thus they won't make as much money. Or that iFixit should be shut down because then explaining how to repair a phone results in repair shops (first or third party) not making as much money.

[StDragon]

6 hours ago, LAwLz said:

I see that as a good thing. The end goal of humanity should be that nobody has to work. That everyone has been replaced by robots.

That's what you hope. But that's not what will happen.

 

When fantasy and reality collide, reality wins out every time.

[LAwLz]

3 hours ago, StDragon said:

That's what you hope. But that's not what will happen.

 

When fantasy and reality collide, reality wins out every time.

I find it interesting that the person pretending to have a crystal ball is talking about "fantasy vs reality".

Since you are so sure about what will happen, can you please enlighten us by sharing your wisdom? 

[Stahlmann]

16 hours ago, TetraSky said:

That's crazy. AI will only ever get better and better, taking jobs away from people.

Who needs fiver for a short video when you can just get an AI to make it even better.

 

It's really unfortunate that no jobs are safe from AI and our robot overlords... Even the oldest job in the world won't be safe for long.

Is it really that bad if AI takes away jobs if it's better at that job than a human would be?

 

There will always be jobs where manual labor or human interference will be needed. Even if AI takes over some jobs, it also creates new jobs. Currently every AI needs a "copilot" for example. And while an AI could probably be trained to plan houses, they still can't build them.

 

12 hours ago, LAwLz said:

The end goal of humanity should be that nobody has to work. That everyone has been replaced by robots.

The big issue is that society might not adapt at a fast enough rate, so we will have some unknown number of years with issues.

2 hours ago, LAwLz said:

I find it interesting that the person pretending to have a crystal ball is talking about "fantasy vs reality".

Since you are so sure about what will happen, can you please enlighten us by sharing your wisdom? 

While this sounds like a noble goal, this is never going to work. History has repeated itself so many times. That should be proof enough that such a utopia is nothing but pure fantasy.

 

The insane amount of stability needed to uphold such a society is impossible to build up in the first place, but holding that stability is even harder when no one has to work and everyone is bored. Just look at Twitter; So much drama is already happening just because people don't have anything better to do or they don't have bigger problems they have to worry about.

 

This whole thought of "nobody has to work, but everyone makes a living" is just so much against the core workings of humanity (not to mention history records) that @StDragon's assumption that this is fantasy is the only one that seems logical.

 

You don't need a crystal ball to see that.

[leadeater]

1 hour ago, Stahlmann said:

There will always be jobs where manual labor or human interference will be needed. Even if AI takes over some jobs, it also creates new jobs. Currently every AI needs a "copilot" for example. And while an AI could probably be trained to plan houses, they still can't build them.

Robots can actually build them, the whole chain of getting that actually done broadly is just more complex and not suited to the today. As for the whole distant future thing sure humans need something to do but you can replace the concept of "job" or "working" for life fulfillment with "hobby" or something else where people can choose to do what they want to do without it being tied to the basic needs for survival aka earning an income to buy food and live etc.

 

Sure it's probably more in the realms of pipedream but we don't actually have to "work", we just need something to do and the more open choice that is the better. Want to spend your life studying classical literature? Go right ahead, at least this currently "dead end" could be done heh.

 

Weighted on balance people give up their passions and dreams to work to survive, that actually is not the best possible life. Neither is pissing around achieving and doing nothing either.

[Needfuldoer]

14 hours ago, LAwLz said:

I see that as a good thing. The end goal of humanity should be that nobody has to work. That everyone has been replaced by robots.

The big issue is that society might not adapt at a fast enough rate, so we will have some unknown number of years with issues.

Problem is, these "AI" models are all trying to automate the kind of creativity you'd expect people to want to pursue, instead of the tedious, menial work that has to be done for society to function but nobody actually wants to do.

 

"TormentNexusBot can compose an opera and paint a masterpiece simultaneously. You go back to stocking shelves and doing your taxes."