You have in all probability seen these mind-blowing Sora movies floating round social media for months. I imply, who hasn’t, particularly now that it’s been formally launched? However here is the factor — Google simply dropped their latest entry to the AI video area, and belief me, it is price listening to.
Due to this information, I’ve spent manner too many hours watching AI-generated movies recently (severely, my YouTube suggestions are a large number), and I’ve acquired to say: the competitors between Sora and Google Veo 2 is getting attention-grabbing. Like, actually attention-grabbing.
However what caught my eye wasn’t simply the flamboyant demos or the technical specs of Google Veo 2, since let’s be trustworthy, these corporations simply cherry-pick one of the best outputs. No, what I cherished seeing was the very same prompts being fed into each techniques. As a result of that is the one technique to really examine these two, proper?
So, no extra ready, let’s dive into this head-to-head comparability.
What’s Google Veo?
Google Veo 2 is the latest model of their video era mannequin, and it is principally what occurs when Google decides to flex its AI muscle mass. We’re speaking a couple of system that may create extremely high-quality movies that look so good, they’ve really outperformed different main fashions in head-to-head comparisons.
What separates Veo 2 from the competitors is that, primary, it’s optimized for filmmaking. Need that excellent dolly shot? Or perhaps that trademark Brian De Palma Break up Diopter shot? Veo 2 will get it. It understands phrases like “18mm lens” or “shallow depth of area” — stuff that administrators and cinematographers take note of.
Quantity two is that, from what I’ve been seeing on-line, Google Veo 2’s expertise is the closest an AI video mannequin has come to understanding actual physics. And one of the best half? They’re rolling this out by means of Google Labs, VideoFX, and even planning to carry it to YouTube Shorts subsequent yr.
What’s Sora?
Introduced in February 2024, Sora is OpenAI’s reply to “what if we might flip textual content into mind-blowing movies?” On the time, it took the AI area by storm as a result of it was probably the most superior text-to-video mannequin we’ve seen up to now. Sora can create reasonable movies as much as 1080p decision and 20 seconds lengthy, full with widescreen and varied facet ratios.
Sora was lastly launched to most of the people in December 2024, which got here free to make use of (for 480p and 720p movies) so long as you’ve gotten a ChatGPT Plus subscription. In the meantime, customers who need higher decision and not using a utilization cap should subscribe to ChatGPT Professional.
Sora vs. Google Veo 2: In contrast
Earlier this yr, I reviewed Sora and concluded that it’s one of the best text-to-video mannequin we’ve seen. Will that assertion nonetheless be true with Google Veo 2 looming over? Let’s discover out.
Instance 1: The Chair Excavation
Credit: @nickfloats
The highest video is from Sora, whereas the underside is from Veo 2.
This was the primary video I’ve seen from Veo 2 and I’m impressed. Clearly, it’s not excellent, but it surely’s already such an enormous step up from a real-world physics standpoint. Sora’s video, whereas good, has that uncanny valley feeling of their motion. To not point out that it doesn’t know the place the chair ends and the sand begins.
Google Veo 2’s model doesn’t have this challenge. Really, aside from one wonky chair, I can’t see any fault within the movies.
Instance 2: Strolling By way of Tokyo
Credit: @nickfloats
The highest video is from Sora, whereas the underside is from Veo 2.
Once more, real-world physics is Sora’s drawback. I really desire the panning movement they applied, however the primary topic (the couple) seems to be taller than different individuals, buildings, and bushes. Additionally they appear to be no-clipping into (phasing by means of) some buildings.
Alternatively, if you happen to confirmed me that Veo 2 video, I wouldn’t have clocked that it got here from AI except I seemed deeper. That’s superb.
Instance 3: The Pirate Battle
Credit: @nickfloats
The highest video is from Sora, whereas the underside is from Veo 2.
This time, I take pleasure in Sora’s output greater than Veo 2. By way of realism, the latter wins. Nevertheless, when it comes to pure creativeness, Sora has the higher idea. It will have been excellent if solely the smaller ship’s bow and stern didn’t swap locations a number of occasions.
Instance 4: Tomato Chopping
Credit: @joecarlsonshow
The highest video is from Sora, whereas the underside is from Veo 2.
Sora’s video someway reduce each fingers and tomato — however neither on the identical time? What within the Schrödinger is going on right here? Clearly, Google Veo 2’s easy however efficient video is the superior output.
Instance 5: Hurdles
Credit: @venturetwins
That is probably the most egregious instance but, in my view. All the things in Sora’s output is unsuitable: the runner is phasing by means of the hurdles, the hurdles itself are the unsuitable type and are both shifting in the direction of or spawning close to the runner, and the topic appears to be like like he’s simply operating in place.
Google Veo 2 continues to be unmistakably AI, however solely due to the non-sensical writing on the hurdles. Apart from that, this might cross as actual sport footage.
Instance 6: A Field of Cash
Credit: @deedydas
Talking of physics, right here’s one other nice instance of Sora missing understanding of real-world physics. When the ball is dropped, it simply retains on going and going — in some unspecified time in the future, the field of cash “explodes” too, which I might see occurring when the ball is dropped, and never whereas it’s within the air.
Google Veo 2 has a greater approximation of gravity. My solely nitpick is that the ball shouldn’t have been completely nonetheless on prime of the cash in the long run, however barely buried.
Instance 7: Gymnastics
Credit: @codebypoonam
The highest video is from Google Veo 2, whereas the underside is from Sora.
I’ll be trustworthy — each of those aren’t sufficient to be mistaken for actual footage. Nevertheless, there’s an enormous distinction in an AI mannequin not figuring out what gymnastics athletes do whereas on air and the way our bodies work. That is one other win for Google Veo 2, little doubt.
When Are We Getting Google Veo 2?
Excellent news for individuals residing within the USA over 18 years previous, you possibly can join and be a part of Google Labs’ waitlist to entry Google Veo 2. Sadly for the remainder of us, we are able to’t use Veo 2 but and there’s no info as to when it’ll be publicly out there.
Given Sora’s announcement to launch date, we are able to perhaps count on Google Veo 2 round early fourth quarter 2025, however that is simply hypothesis.
All Stated and Executed
Google Veo 2 is exhibiting us what occurs when one among tech’s greatest and most revered gamers decides to get severe about AI video era. Whereas Sora blazed the path and nonetheless holds its personal when it comes to inventive interpretation, Veo 2’s understanding of real-world physics is simply… on a unique degree.
Is it excellent? Nope. Does it nonetheless have that occasional “AI weirdness”? You wager. But when we’re speaking about which one’s nearer to creating movies that might cross for actual footage — Veo 2 takes the crown. And that is coming from somebody who was completely blown away by Sora just some months in the past.
However the true winner right here is us: the customers. As a result of competitors breeds innovation, and with these Google and OpenAI pushing one another to do higher, we’re about to see some severely spectacular stuff within the video AI area.
Simply bear in mind — we’re nonetheless within the early days. But when that is what AI video era appears to be like like now, think about the place we’ll be this time subsequent yr. Thrilling occasions forward!
