You’ve probably seen those mind-blowing Sora videos floating around social media for months. I mean, who hasn’t, especially now that it’s been officially released? But here’s the thing — Google just dropped their newest entry to the AI video space, and trust me, it’s worth paying attention to.
Because of this news, I’ve spent way too many hours watching AI-generated videos lately (seriously, my YouTube recommendations are a mess), and I’ve got to say: the competition between Sora and Google Veo 2 is getting interesting. Like, really interesting.
But what caught my eye wasn’t just the fancy demos or the technical specs of Google Veo 2, since let’s be honest, these companies just cherry-pick the best outputs. No, what I loved seeing was the exact same prompts being fed into both systems. Because that’s the only way to actually compare these two, right?
So, no more waiting, let’s dive into this head-to-head comparison.
What is Google Veo?
Google Veo 2 is the newest version of their video generation model, and it’s basically what happens when Google decides to flex its AI muscles. We’re talking about a system that can create incredibly high-quality videos that look so good, they’ve actually outperformed other leading models in head-to-head comparisons.
What separates Veo 2 from the competition is that, number one, it’s optimized for filmmaking. Want that perfect dolly shot? Or maybe that trademark Brian De Palma Split Diopter shot? Veo 2 gets it. It understands terms like “18mm lens” or “shallow depth of field” — stuff that directors and cinematographers pay attention to.
Number two is that, from what I’ve been seeing online, Google Veo 2’s technology is the closest an AI video model has come to understanding real physics. And the best part? They’re rolling this out through Google Labs, VideoFX, and even planning to bring it to YouTube Shorts next year.
What is Sora?
Announced in February 2024, Sora is OpenAI’s answer to “what if we could turn text into mind-blowing videos?” At the time, it took the AI space by storm because it was the most advanced text-to-video model we’ve seen so far. Sora can create realistic videos up to 1080p resolution and 20 seconds long, complete with widescreen and various aspect ratios.
Sora was finally released to the general public in December 2024, which came free to use (for 480p and 720p videos) as long as you have a ChatGPT Plus subscription. Meanwhile, users who want better resolution without a usage cap must subscribe to ChatGPT Pro.
Sora vs. Google Veo 2: Compared
Earlier this year, I reviewed Sora and concluded that it’s the best text-to-video model we’ve seen. Will that statement still be true with Google Veo 2 looming over? Let’s find out.
Example 1: The Chair Excavation
The top video is from Sora, while the bottom is from Veo 2.
This was the first video I’ve seen from Veo 2 and I’m impressed. Obviously, it’s not perfect, but it’s already such a huge step up from a real-world physics standpoint. Sora’s video, while good, has that uncanny valley feeling in their movement. Not to mention that it doesn’t know where the chair ends and the sand begins.
Google Veo 2’s version doesn’t have this issue. Actually, apart from one wonky chair, I can’t see any fault in the videos.
Example 2: Walking Through Tokyo
The top video is from Sora, while the bottom is from Veo 2.
Again, real-world physics is Sora’s problem. I actually prefer the panning motion they implemented, but the main subject (the couple) appears to be taller than other people, buildings, and trees. They also seem to be no-clipping into (phasing through) some structures.
On the other hand, if you showed me that Veo 2 video, I wouldn’t have clocked that it came from AI unless I looked deeper. That’s amazing.
Example 3: The Pirate Battle
The top video is from Sora, while the bottom is from Veo 2.
This time, I enjoy Sora’s output more than Veo 2. In terms of realism, the latter wins. However, in terms of pure imagination, Sora has the better concept. It would have been perfect if only the smaller ship’s bow and stern didn’t swap places multiple times.
Example 4: Tomato Cutting
The top video is from Sora, while the bottom is from Veo 2.
Sora’s video somehow cut both fingers and tomato — but neither at the same time? What in the Schrödinger is happening here? Clearly, Google Veo 2’s simple but effective video is the superior output.
Example 5: Hurdles
This is the most egregious example yet, in my opinion. Everything in Sora’s output is wrong: the runner is phasing through the hurdles, the hurdles itself are the wrong form and are either moving towards or spawning near the runner, and the subject looks like he’s just running in place.
Google Veo 2 is still unmistakably AI, but only because of the non-sensical writing on the hurdles. Other than that, this could pass as real sport footage.
Example 6: A Box of Coins
Speaking of physics, here’s another great example of Sora lacking understanding of real-world physics. When the ball is dropped, it just keeps on going and going — at some point, the box of coins “explodes” too, which I could see happening when the ball is dropped, and not while it’s in the air.
Google Veo 2 has a better approximation of gravity. My only nitpick is that the ball shouldn’t have been perfectly still on top of the coins in the end, but slightly buried.
Example 7: Gymnastics
The top video is from Google Veo 2, while the bottom is from Sora.
I’ll be honest — both of these aren’t enough to be mistaken for real footage. However, there’s a big difference in an AI model not knowing what gymnastics athletes do while on air and how bodies work. This is another win for Google Veo 2, no doubt.
When Are We Getting Google Veo 2?
Good news for people living in the USA over 18 years old, you can sign up and join Google Labs’ waitlist to access Google Veo 2. Unfortunately for the rest of us, we can’t use Veo 2 yet and there’s no information as to when it’ll be publicly available.
Given Sora’s announcement to release date, we can maybe expect Google Veo 2 around early fourth quarter 2025, but this is just speculation.
All Said and Done
Google Veo 2 is showing us what happens when one of tech’s biggest and most respected players decides to get serious about AI video generation. While Sora blazed the trail and still holds its own in terms of creative interpretation, Veo 2’s understanding of real-world physics is just… on a different level.
Is it perfect? Nope. Does it still have that occasional “AI weirdness”? You bet. But if we’re talking about which one’s closer to creating videos that could pass for real footage — Veo 2 takes the crown. And that’s coming from someone who was absolutely blown away by Sora just a few months ago.
But the real winner here is us: the users. Because competition breeds innovation, and with these Google and OpenAI pushing each other to do better, we’re about to see some seriously impressive stuff in the video AI space.
Just remember — we’re still in the early days. But if this is what AI video generation looks like now, imagine where we’ll be this time next year. Exciting times ahead!
