OpenAI Just Changed Video Forever
OpenAI has recently unveiled Sora, an AI text-to-video model that has completely revolutionized the way we look at AI-generated videos. With the ability to generate videos up to 1 minute long, Sora showcases an unprecedented level of detail and coherence that has never been seen before.
From a stylish woman walking down a Tokyo street to a movie trailer featuring the adventures of a 30-year-old spaceman, Sora’s capabilities are truly mind-blowing. The AI model can even generate stunning drone footage of an old California mining town, capturing intricate details without any distortion.
One of the most impressive features of Sora is its ability to generate natural-looking eye movements and blinking. The model also utilizes a clever vignetting trick to focus on center frame details, ensuring a more realistic output.
While Sora is not yet available to the public, OpenAI is currently seeking feedback from Red teamers, artists, designers, and filmmakers to further fine-tune the model. With Sora, the possibilities for AI-generated videos are truly endless, opening up a new realm of creativity and innovation in the world of video production.
Watch the video by Theoretically Media
Video Transcript
Hey everyone so I was not planning on making a video today which are famous last words in AI uh because open AI just dropped the mic on AI video open AI are showing off Sora which is an AI text to video model um that is completely insane
It generates videos up to 1 minute long and the level of detail and coherency is something that we have not seen yet so this is pretty nuts this is a stylish woman walking down a Tokyo Street warm glowing Neon animated City signage there’s a bit of a stumble in the walk
Cycle but I mean I can totally forgive that considering how smooth the rest of the motion is and the fact that she’s staying super super coherent also I’m going to just jump ahead just a little bit the fact that as it continues on there’s a cut in here as well which is
That’s it’s pretty it’s pretty wild uh it looks really really great it’s the same character obviously as well uh speaking of cuts this one I mean just blew me away this is a movie trailer featuring The Adventures of a 30-year-old Spaceman wearing a red wool knitted motorcycle helmet handheld
Footage Cuts this looks like it could be an actual filmed trailer I mean that there’s nothing really in here to majorly indicate that this is AI generated here’s some really gorgeous Drone footage as well this is a bit of a shorter clip I think this one only runs
8 seconds but even then with current video generators this would have to be an extend and obviously once you get past that 3 or 4 second Mark things tend to start to fall apart uh Sora does not seem to be falling apart at all so it is
Important to note Sora is not available yet it is instead being seeded out to Red uh teamers to assess for critical areas of harms or risk and it’s also going out to a number of artists designers and filmmakers to sign test out to see what it can do and basically
Provide feedback to open AI but like this Drone footage of an Old California Mining town this looks looks really really pretty great um you’ve got all of those buildings A lot of people walking around on horses they aren’t really morphing out and going insane and even
As the camera makes this turn here the buildings stay intact they don’t start to shift and warp and morph into weird things although I did notice one small incoherency um as it’s coming along here we get this like weird icon that’s there I do not think that that is period
Specific that’s where the gold is I moov looks incredible in Sora uh this is definitely the most natural looking eye movement and blinking I have seen yet I did notice one kind of thing reviewing a lot of this footage and that Sora seems to prefer this trick where it kind of
Vignettes out on the sides um that’s actually pretty smart because it probably means that it does the model doesn’t have to pay attention to as much detail across the frame if it kind of blurs things out a little bit it probably has more resources to concentrate on Center frame that is just
A theory though movement is really incredible I’m guessing this has to be running at 24 frames a second or really close to it uh and indeed this is the best dancing generation that I’ve seen outside of video to video um yeah this is really really quite good here’s some
More footage that completely blew me away I mean there’s a lot of people and colors and movement and things happening in this scene and the fact that everything is staying relatively coherent is pretty wild I mean uh especially as we pull out and we now got
This giant City Scene now I did notice in this shot that um the boys hands do get a little bit AI wacky Fingers um here around the 11 second Mark but I mean considering everything else in this sequence it’s it’s pretty minor I love this piece as well it’s like a 20 second
Shot of you know these various old CRT TVs uh playing old movies uh yeah this is there’s a lot going on here I would say that our current video generators probably can’t do this you would either end up with a lot more blank TVs static TVs or just scenes of people that are
Still uh there’s a lot of movement happening in here this Range Rover driving sequence is by far the best driving that I have seen yet in any AI video um you know the Range Rover here traveling down the mountain road is bumping it definitely looks like it has
Weight it’s staying on the road and the path as well um yeah this is pretty impressive open AI notes that Sora is able to generate complex scenes with multiple characters and specific types of motion uh with accurate details on the subject in the background uh definitely kind of showcased here in
This Chinese Lunar New Year celebration that legitimately looks like it could be like news broadcast footage here’s kind of a steady cam shot going through a museum hallway I always appreciate these kinds of videos as well considering that the model has to populate the art for
The room itself but yeah each of these pieces also stays you know temporarily coherent within the scene as well like basically it knows enough not to have the subjects of the painting begin moving around the model does seem to have a pretty good amount of imagination
As well the prompt here is a robot’s life in a cyberpunk city I will say that the backgrounds do feel a little bit on the static side but the fact is that the robot actually stays consistent throughout each of these shots as well uh level of detail on them is pretty
Remarkable as well yeah this looks really really pretty cool here’s another one of an old guy pondering existence in a Parisian Cafe that’s not me making fun of the video that’s actually the prompt um although it definitely sells that’s what this dude is doing this looks so
Good I mean I think that if you were to pause on any frame it basically looks like a mid Journey you know V5 or V6 image here’s some shots of puppies playing in the snow because puppy playing in the snow um yeah this I mean adorable it’s AI generated or not this
Is adorable uh just watch out for that yellow snow now I do appreciate appreciate the fact that open AI actually provided a number of uh you know those weird AI video mistakes that Sora makes just letting everyone know like look things still get weird which I
Appreciate um so yeah here we have a man running backwards on a treadmill and look I’m on record as saying that you know once all of the bugs are actually Stamped Out of AI video I’m actually going to miss our current surreal area I mean stuff like this it genuinely cracks
Me up I mean I know it can get frustrating when you’re looking for something very specific but at the same time it’s it’s good for a laugh and I’ll say at the video length that SAR is providing I think that even if things get weird like this woman’s hand here
Kind of like freaks out on us there’s enough in the length of the shot that you can probably pull something usable out of it as a note open AI will be utilizing ctpa metadata in the Sora video output it’s basically you know code in there that will indicate that
This is indeed AI generated video and honestly I mean as good as that looks I have no problems with that there will also obviously be guard rails put on the model should and when it is released uh basically you know preventing hateful imagery celebrity likeness and IP it’s totally understandable especially
Considering how good Sora is looking another shot of a man sitting on a floating cloud reading a book this I mean it really looks like it could come out of like an Old Navy or a gap ad so I am really excited to learn about the technical details when AI releases that
Later or at least understanding it to you know the ability that I have uh but it is interesting to note that you will be able to do uh image to video with this model as well so obviously there’s no model to play with yet but if
Anything I think this does give us a really good indication of what we’re going to be seeing in AI generated video within this year whether this be coming out of open AI or another company this is what things are going to be looking like very soon so yeah just a quick
Video on this again I wasn’t even planning on making a video today but between this and the boxes earlier this week this has been a crazy week for AI video I’ll definitely be keeping an eye on Sora and I will let you know as details develop I thank you for watching
My name is Tim
Video “OpenAI Just Changed Video Forever” was uploaded on 02/16/2024 to Youtube Channel Theoretically Media