Exploring the Future of AI: A Comprehensive Review of OpenAI’s Sora Video Generation Tool
This ones ai-generated too okay what about this one pay really close attention did you catch it see no okay that one was real but the fact that you thought about it so i for the last week have had full access to the newest and most advanced version of sora which is the ai video generation tool from openai and the results that i’ve gotten from it are both horrifying and inspiring at the same time the last time i looked at this nine months ago this was a very private tool that was just unveiled they were just starting to show its capabilities and so we were looking at other peoples prompts actually basically it was openai’s carefully hand-selected prompts for what they wanted to show us and we could learn some stuff from that but this time i just get the controls myself its just an open text box my wish is soras command and that’s actually kind of a lot of pressure you can type whatever you want but i did not take this responsibility lightly.
I basically over the past couple days have been taking every possible angle of asking sora to make things for me I’ve asked for photorealistic things I’ve asked for cartoon things I’ve asked for objects people signs actions text still life everything you can imagine and i feel like I’ve come away with a pretty good sense of what its good at what its bad at and what it could actually be used for today but this is a powerful tool that’s about to be in the hands of millions of people so consider this the first ever sora review the good the bad and the ugly lets jump in so first of all this is the UI on the left side you can explore other prompts that people have recently made and then some featured ones where i suppose openai is still gonna curate some of the best results to showcase what it can do if you hit the bookmark button on any of these they’ll show up in your saved tab and what i like about these again is anything you click on you can see obviously.
who made it but you can also read exactly what they typed into the prompt box to get this result and ill get back to that in a second so then underneath that you have your library of all the videos I’ve generated from sora prompts anything I’ve tagged as a favorite from my own creations and then uploads which is any of the files that I’ve uploaded to sora for it to make videos out of another crazy feature that ill get to in a second but then lastly you have your folders so you can organize things that you’ve created into different folders maybe for different projects or different themes or whatever just to keep it somewhat organized so okay right off the bat sora is a tool after all.
So this feature of being able to see what other people are typing and then the results that they create from those prompts is both for inspiration but also for education because it is really interesting reading how simple or how detailed some of these prompts are and then looking at what sora generated and what it took creative liberties with to add on its own and then if you really like someone else’s results but wanna adjust it in just a slight way for your own use there’s the remix feature so you hit that button and then you can describe specific changes you want it to make to that video for it to then generate a new one so for this one for example i like this shot and the house is pretty cool but i kind of want there to be a golf course on those cliffs in the background right so i type that and then i can also change the resolution which as you can see.
It will take longer for higher resolutions and then you also can specify exactly how much of a remix it is so if its a subtle mild or strong remix which dictates basically how much its willing to change the results from the original and then you can even dial that in on a slider from one to eight if you want to i think adding golf course would be nice but i wanna keep those cliffs and the house so its a mild change right click remix wait a bit and then boom you have the new creation a fresh new artificially-generated video with that same house and those same cliffs but now 1080p and with a golf course in the background wild.
I’ve noticed its hard to get a video to do several different things in succession with a single prompt but you can use their built-in editor to string several prompts together in a row and then storyboard can help you blend them all together into one longer video and that actually can work better instead.
so I’ve been playing with this all week and I’ve been throwing dozens and dozens and dozens of prompts at it with all sorts of characters different styles different unique things in the video and so here are some observations that may surprise you or maybe not so one it seems like there is no object permanence so we had ai-generated images right and they’ve gotten you’ve seen the ark of them getting better and better and higher resolution and realistic over time now that were doing videos basically they’re stringing together a bunch of different ai-generated photos in a row with some sort of continuity but for a video.
A human the objects kind of have to move in certain ways and behave in certain ways in the real world and so one of the most common dead giveaways from a flaw of an ai-generated video especially with this tool is just errors in object permanence so you’ll see things like objects passing in front of each other or behind each other in ways that don’t make sense you’ll see stuff disappear or reappear out of thin air especially if objects pass in front of them but even sometimes without any real reason at all they can also just materialize and then vanish like the smartphone in this imaginary tech reviewers hands bop just gone and there’s also another one that’s really common anytime something with legs has to walk.
The Evolution of Soft Electronic Music: Real vs. AI-Generated Sounds in Today’s Digital Landscape
Real or ai-generated soft electronic music okay that ones ai-generated not a lot of giveaways unless you happen to have memorized the geographical landscape around mount fuji and hunted for inconsistencies but okay what about this one and yeah you guessed it already this ones ai-generated too okay what about this one pay really close attention did you catch it see no okay that one was real but the fact that you thought about it so i for the last week have had full access to the newest and most advanced version of sora which is the ai video generation tool from openai.
The results that l’ve gotten from it are both horrifying and inspiring at the same time the last time i looked at this nine months ago this was a very private tool that was just unveiled they were just starting to show its capabilities and so we were looking at other peoples prompts actually basically it was openai’s carefully hand-selected prompts for what they wanted to show us and we could learn some stuff from that but this time i just get the controls myself its just an open text box my wish is soras command and that’s actually kind of a lot of pressure.
You can type whatever you want but i did not take this responsibility lightly I basically over the past couple days have been taking every possible angle of asking sora to make things for me I’ve asked for photorealistic things I’ve asked for cartoon things I’ve asked for objects people signs actions text still life everything you can imagine and i feel like I’ve come away with a pretty good sense of what its good at what its bad at and what it could actually be used for today but this is a powerful tool thats about to be in the hands of millions of people.
The Art of Deception: How CCTV-Style Footage Can Mislead with CGI
Since it doesn’t know that they’re objects sometimes also doesn’t really make sense pretty much anytime i try anything remotely photorealistic basically you instantly notice that the movements don’t look right it tends to look kind of slow motion actually but then other parts wont be in slow motion which our eye picks up on right away i was playing with making like cctv-style footage cause in my head that’s gonna be the easiest thing to fake people are gonna get their hands on this they’re gonna fake security camera footage.
All the time but even in this you can see the movement of the humans is just slightly off first its a little slow then its fast for no reason its just weird now it seems to be decent sometimes at fluid dynamics for whatever reason I’ve been very impressed with some examples of water rippling or crashing in waves or moving in realistic waves it does happen sometimes it can be actually pretty good-looking and same thing also with fire sometimes it can look pretty realistic even if the smoke isn’t but yeah physics in general is definitely a weakness so then if you know this you can mess more with Claymation or cartoon style and then those irregularities and movement or physics become a lot easier to palate because it feels more like a stylistic choice or an artistic choice but ill get to the stuff that it actually does really well in a second but there is another feature.
Exploring Sora: A Comprehensive Review of the Good, the Bad, and the Ugly
So consider this the first ever sora review the good the bad and the ugly lets jump in so first of all this is the UI on the left side you can explore other prompts that people have recently made and then some featured ones where i suppose openai is still gonna curate some of the best results to showcase what it can do and if you hit the bookmark button on any of these they’ll show up in your saved tab and what i like about these again is anything you click on you can see obviously who made it but you can also read exactly what they typed into the prompt box to get this result and ill get back to that in a second so then underneath that you have your library of all the videos .
I’ve generated from sora prompts anything I’ve tagged as a favorite from my own creations and then uploads which is any of the files that I’ve uploaded to sora for it to make videos out of another crazy feature that ill get to in a second but then lastly you have your folders so you can organize things that you’ve created into different folders maybe for different projects or different themes or whatever just to keep it somewhat organized so okay right off the bat sora is a tool after all and so this feature of being able to see what other people are typing and then the results that they create from those prompts is both for inspiration but also for education because it is really interesting reading how simple or how detailed some of these prompts are and then looking at what sora generated and what it took creative liberties with to add on its own and then if you really like someone else’s results but wanna adjust it in just a slight way for your own use there’s the remix feature so you hit that button and then you can describe specific changes you want it to make to that video for it to then generate a new one.
So for this one for example i like this shot and the house is pretty cool but i kind of want there to be a golf course on those cliffs in the background right so i type that and then i can also change the resolution which as you can see it will take longer for higher resolutions and then you also can specify exactly how much of a remix it is so if its a subtle mild or strong remix which dictates basically how much its willing to change the results from the original and then you can even dial that in on a slider from one to eight if you want to i think adding golf course would be nice but i wanna keep those cliffs and the house so its a mild change right click remix wait a bit and then boom you have the new creation a fresh new artificially-generated video with that same house and those same cliffs but now 1080p and with a golf course in the background wild.
I’ve found that 360p videos take actually very little time to generate i do not want to use the word create here these are artificially-generated from source material but basically a five-second 360p video will generally take less than 20 seconds to generate and they’re being generated on openai servers so it has nothing to do with the speed of my computer or even my internet connection once i send off the prompt it just takes that time and then when its done it shows up and i can download.
Exploring the Evolution of AI-Generated Videos: Insights on 1080p Quality and Object Permanence
It if i want to and then something like a 1080p 10-second video it looks significantly better but it also takes much longer it can take a couple of minutes for a really detailed prompt and a 1080p video and that’s also right now when almost no one else is using it i kind of wonder how much longer it’ll take when this is just open for anyone to use but that’s roughly how long they’ve taken and there’s also the storyboard feature which kind of looks and acts like an online video editor where you can string together several prompts and it works best for stringing together several different actions in a row basically I’ve noticed its hard to get a video to do several different things in succession with a single prompt but you can use their built-in editor to string several prompts together in a row and then storyboard can help you blend them all together into one longer video and that actually can work better instead so I’ve been playing with this all week.
I’ve been throwing dozens and dozens and dozens of prompts at it with all sorts of characters different styles different unique things in the video and so here are some observations that may surprise you or maybe not so one it seems like there is no object permanence so we had ai-generated images right and they’ve gotten you’ve seen the ark of them getting better and better and higher resolution and realistic over time.
Now that were doing videos basically they’re stringing together a bunch of different ai-generated photos in a row with some sort of continuity but for a video to make sense to a human the objects kind of have to move in certain ways and behave in certain ways in the real world and so one of the most common dead giveaways from a flaw of an ai-generated video especially with this tool is just errors in object permanence so you’ll see things like objects passing in front of each other or behind each other in ways that don’t make sense.
You’ll see stuff disappear or reappear out of thin air especially if objects pass in front of them but even sometimes without any real reason at all they can also just materialize and then vanish like the smartphone in this imaginary tech reviewers hands bop just gone and there’s also another one that’s really common anytime something with legs has to walk if you just watch the legs for long enough they almost guaranteed will mess up which leg is the front leg and which is the back just look at this one just try to watch one of the legs and you’ll see it happens it switches back and forth from the front leg to the back leg.
This is super common it happens all the time in these and that brings me to another one of my biggest observations which is it just struggles with physics in general the way things move which actually makes perfect sense because physics would seem to require some understanding of what you’re making a video about so in the same way that large language models can struggle with hallucinations or putting together sentences that are technically incorrect or don’t quite make sense these video models will also struggle with putting together videos where the movement of the objects.
Understanding the Challenges of Video Models: Why Leg Movement Often Confuses Front and Back
If you just watch the legs for long enough they almost guaranteed will mess up which leg is the front leg and which is the back just look at this one just try to watch one of the legs and you’ll see it happens it switches back and forth from the front leg to the back leg this is super common it happens all the time in these and that brings me to another one of my biggest observations which is it just struggles with physics in general the way things move which actually makes perfect sense because physics would seem to require some understanding of what you’re making a video about so in the same way that large language models can struggle with hallucinations or putting together sentences.
That are technically incorrect or don’t quite make sense these video models will also struggle with putting together videos where the movement of the objects since it doesn’t know that they’re objects sometimes also doesn’t really make sense pretty much anytime i try anything remotely photorealistic basically you instantly notice that the movements don’t look right it tends to look kind of slow motion actually but then other parts wont be in slow motion which our eye picks up on right away i was playing with making like cctv-style footage cause in my head that’s gonna be the easiest thing to fake people are gonna get their hands on this they’re gonna fake security camera footage all the time but even in this you can see the movement of the humans is just slightly off first its a little slow.
Then its fast for no reason its just weird now it seems to be decent sometimes at fluid dynamics for whatever reason i’ve been very impressed with some examples of water rippling or crashing in waves or moving in realistic waves it does happen sometimes it can be actually pretty good-looking and same thing also with fire sometimes it can look pretty realistic even if the smoke isn’t but yeah physics in general is definitely a weakness so then if you know this you can mess more with Claymation or cartoon style and then those irregularities and movement or physics become a lot easier to palate because it feels more like a stylistic choice or an artistic choice but ill get to the stuff that it actually does really well in a second but there is another feature that you might have seen around the internet a little bit with some other tools like this that takes a source image and can turn it into a video with a prompt so people have been bringing memes to life and imagining what would happen right after some iconic image just by typing into the ai video generator and asking sora what it would look like and sora can do that too the difference with sora i’ve found is its way less likely to try anything.
That it recognizes as having any copyrighted or intellectual property whatsoever its pretty picky it actually does reject a lot of stuff anything with any public figures or recognizing characters in general or logos it actually refuses to do it it will also refuse if it thinks any subject you upload in the image is under 18 years old makes sense but i did upload this ai-generated image which dolly made and asked it to make them all singing and dancing and hit generate and it totally did it again it helps a lot that they are cartoon styled and not some super realistic scene but this is a totally ai-generated video i also tried a few other images from my camera roll and these just it just gets weird it just doesn’t know the context of what direction any object was moving in the photo again it doesn’t know physics and so things just get really wonky its impressive that its ai-generated video but you can tell pretty quickly that its ai-generated video so all that being said we’ve seen what its good at we’ve seen what its not so good at and you might be wondering what is this tool actually for and i actually found a few things that are not so scary that are actually pretty good.
These are the things that i found as of right now that sora can be used for abstracts so you can be as descriptive as you want and create all sorts of textures and colors and gradients to make an abstract shape move around in a way that essentially looks like it could be a screensaver or just some background piece or whatever you want i’m sure there’s a world where someone’s gonna make an net out of this stuff also totally weird but you can do abstract stuff for sure i noticed its getting better at reproducing text especially when you ask it for specific text so there are sometimes background garbled text in these videos and that’s gonna continue to happen but i’ve found that if i asked it for an individual title slide in a style it actually could give me the correct text on that slide this animation it gave me of sketching the empire state building this would make a killer title slide or even an intro to a documentary about the building or Manhattan or something like that i was very impressed and then stop motion or cartoon-style characters again like i said before cartoons don’t necessarily have to have realistic movements or physics so the errors in those things don’t hit the eye as much and so the characters still look like dare i say art they’re characters especially in scenes like this one the graffiti doesn’t have to actually say real words for this to look normal the bear doesn’t have to have the correct number of fingers or toes it just looks like some album art come to life basically which i’m sure someone would find useful on social media.
Somewhere and while were at it here’s a couple other random prompts that impressed me i asked for santa fighting frosty the snowman in the style of mortal Kombat and it built a readable scoreboard i didn’t ask for the scoreboard but it built all that with the names in everything which is pretty crazy immediately again the details are weird when you start looking at it but this could be a good inspirational starting point and then i also asked for a video of a tech reviewer sitting at a desk and i said nothing about this fake plant but sora took the liberty of adding this exact fake plant here which i said nothing about in my prompt and that is a convenient reminder of the other side of the tools like this the unknown behind a lot of the tools clearly we already know that these are ai-generated videos that have to come from some source material are my videos in that source material is this exact plant part of the source material is it just a coincidence i don’t know openai has talked about using publicly available media and data.
Its always to me been kind of a sketchy definition we don’t know if its everything we don’t know if its too late to opt out if we wanted to opt out and also i have no idea how much energy this uses or if it is significantly more than the ll’ms or dolly were using but still the craziest part of all of this is the fact that this tool sora is going to be available to the public i guess around the time of this video publish to everyone to millions of people all at once and i mean yes it does a pretty good job with guardrails of refusing to do anything photorealistic refusing to use actual peoples likenesses or any dangerous or harmful acts and yes it does watermark every single generated video with this little animation here in the corner but you can still crop watermarks and its still an extremely powerful tool that directly moves us further into the era of not being able to believe anything you see online i mean read through the comments on my last video about ai-generated video this is a lot for humanity to digest right now all of that and these videos don’t even have audio they’re just a few seconds long they’re just 1080p this is the new baseline this is once again the worst that they will be as they continue to get better in the future its a lot to think about
