Real life or AI? Watch these videos and see if you can spot the difference

Having spent some time with generative AI, I thought I had a fair idea of what to expect from Veo 3 — Google’s cutting-edge AI video generator. But when I finally ponied up the $20 for a Google AI Pro subscription a few weeks ago, I was surprised to find that it outperformed even my most optimistic expectations. Unlike early AI image generators that would produce obvious deformities like extra fingers or absurd architecture, Google’s Veo 3 can generate videos that look strikingly similar to their real world equivalents.

In fact, some of Veo’s videos can look so convincing on social media that I had to double-check whether I was looking at AI-generated content or a stock clip. Naturally, that led to the question: how good is Veo 3, really — and could the average person even tell that they’re looking at an AI-generated video? To find out, I’ve put together a short quiz below with six Veo-generated clips pit against real-world videos. Can you tell the difference?

Veo 3’s ability to generate extremely convincing clips is impressive in its own right, but it also goes one step further: it can also produce synchronized speech or sound effects. This means the results it produces can seem nearly indistinguishable from the real deal to the untrained eye.

Of course, there are telltale signs pointing to a synthetic video’s AI origins if you look closely but you can expect those minor imperfections to be gone sooner rather than later. Google has already dispatched numerous fixes to Veo 3 since its debut at I/O, including a recent one that prevents glitchy subtitle-like text from appearing.

To generate a video using Veo 3, you will need a Google AI Pro or Ultra subscription. That will set you back a minimum of $20 per month, to speak nothing of the higher tier that costs at an eye-watering $250 each month. And even then, you only get a limited amount of generation credits per month.

The list of Veo 3 limitations doesn’t end there. You can only generate extremely short videos at the moment — no longer than eight seconds each. That said, Google Flow, an experimental AI filmmaking tool, allows you to chain multiple Veo-generated clips together to create a longer video. Length aside, the other big limitation is that you can only generate 720p videos with Veo 3 at the moment.

Veo 3 costs Google a lot of money in terms of processing — and while we don’t know the exact internal cost to Google, we do know what developers are charged to use Veo 3 via an API. Each second of video with audio costs $0.75 to generate, while silent clips cost $0.50 per second. That means an 8-second video costs developers up to $6 per generation. Multiply that by just a few clips and it becomes clear why Google limits how many generations you get with a $20 Pro subscription. The cost of this tech is likely far from trivial.

So, is Veo 3 worth that princely price tag? That brings us back to the original question: can you actually tell the difference between a real world video and an AI-generated one? Below, I’ve lined up six short clips — Let’s see if you can spot which is which.

Let’s start with an easy one. This one’s relatively simple to pick out if you’re looking closely. The AI-generated version doesn’t replicate many of the real-world details you’d expect in a genuine farming scene. The sky, farm machinery, and smaller background elements look a little too clean and uniform. But to be fair, I gave Veo 3 a pretty short and non-descriptive prompt.

Considering that, Veo 3 actually did an excellent job. If you weren’t looking at the video side by side with real footage, it could easily pass for the real thing at a glance. What’s more impressive is that I asked for a specific machinery color scheme and even mentioned the brand name, and Veo 3 delivered on both fronts. That shows just how good this model is at following context and direction — even if it doesn’t nail the finer details just yet.

Another relatively easy one. While Veo 3’s version comes impressively close, especially with the subtle body movements and surprisingly convincing ambient sound, it falls short when placed next to real stock footage. The AI squirrel looks just a bit too clean, and the background is too dark — although my prompting could be to blame. The most impressive part, though? I instructed Veo 3 to focus on the squirrel’s fur with a shallow depth of field, and it delivered.

I think what gives it away is the lack of any unpredictable authenticity you get with real animals. In the stock clip, the squirrel fumbles with the nut, bites off more than it can chew (literally), and has a bit more character. Still, if you saw the AI clip on its own, you’d probably never question it.

Veo 3 shows off its strengths here, nailing the overall atmosphere — the bustling energy and sense of movement. If you’ve never been to Thailand, both videos might look equally convincing.

But look closer and the cracks begin to show. The stalls are too uniform and lack the visual clutter that you’d see in a real night market. The vendors also seem to be selling random, mismatched items that don’t make much sense side by side. And if you look at the vendors’ hand movements, you’ll see that they’re rather unnatural. This is a classic telltale sign of generative AI, and Google’s video generator is not immune to that problem.

Still, this is a difficult scene to pull off, and considering the complexity, Veo 3’s attempt is half decent.

This scene is perhaps the most impressive of the bunch. Without the clutter of city elements or complex character interactions, Veo 3 can truly shine. Even with dramatic lighting, scenic landscapes, and atmospheric effects like fog, it doesn’t really break a sweat. It helps that the real-world clip looks striking too, almost like something out of a video game.

That makes this one genuinely tricky to guess. Need a hint? Look closely at the hiker’s left hand and you’ll notice a subtle rendering hiccup that breaks the illusion.

Another difficult one. Veo 3 delivers an impressive result here, and at first glance, it’s genuinely hard to tell the AI-generated video apart from the real thing. The pacing and the motion of the goats look convincing enough.

I don’t know if I would be able to tell them apart, but knowing which one is AI-generated, I can pick out subtle oddities. For example, the ground in the AI clip feels just a bit too flat. The goats’ faces and bodies are also strangely smooth, whereas the real animals have some grime on them. Still, there’s no single glaring flaw — it’s more of a gut feeling.

Some of the above clips were easier to detect than others, but if you found yourself second-guessing even the obvious ones, you’re not alone. When AI-generated videos get the lighting, camera angles, and subjects mostly right, it can become surprisingly difficult to spot. I’m not sure I would’ve picked up on many of the fakes without a direct comparison, even though I’ve looked at hundreds or thousands of AI-generated images.

As the technology becomes cheaper, you can expect videos made using Veo 3 to become more commonplace. Google currently adds a small watermark to the bottom-right corner of all AI-generated videos, but if you didn’t notice it above, that’s because I cropped it out of every single clip. Doing that took all of a few minutes per video, meaning we need to find a new and more effective way of dealing with the impending deluge of fake videos on the Internet. I don’t know what the solution is, but I hope Google’s AI ethics team does.