Has *Watch Dogs* Entered Reality? AI Face Swapping Has Reached the Millisecond Level

What sent a chill down my spine wasn’t that the image had become clearer, but that it had become almost as fast as a real-time online connection. In the past, you could still rely on a bit of latency to sense that something was off. Now, the information given in the CCTV report was very direct: AI face-swapping has already achieved millisecond-level processing. From inputting a face to generating the replacement, it’s almost done in the blink of an eye. The feeling is a bit like in Watch Dogs, where the system seamlessly takes over a real identity—except this time, what’s being hacked isn’t a server, but our eyes.

Speaking of which, what truly surprised me is that the barrier to entry is still dropping. In the demonstration, the system only needed a single static photo to swap the face of a person in a video with the target person in real time. After the swap, it’s not just that the facial features look similar—skin texture and lighting details can also be mapped on convincingly, and dynamic movements like head turns and blinking are tracked very steadily. In the past, this kind of effect still felt like a rough meme clip. Now it already has a bit of that Return of the Obra Dinn kind of “you think you understand what you’re seeing, but all the clues are wrong” sense of disorientation.

Interestingly, the “multiple avatars” mentioned in this report does not mean sci-fi-style cloning of several real people, but rather generating multiple highly realistic digital faces in a video, making it seem as if the same person can appear in different scenes at the same time. This kind of visual impact will further erode ordinary people’s judgment. If voice cloning is added on top, with the voice, expressions, and lip movements all matching up, then trying to tell the difference with just your eyes and ears is basically no longer realistic.

But unfortunately, the faster the technology advances, the more immediate the risk of abuse becomes. The most typical scenario right now is impersonating relatives or friends in a video call to trick people into transferring money or lending money. Police have already issued multiple notices confirming that some people were indeed scammed out of large sums after receiving video calls from “people they knew.” On the other hand, it could also be used to bypass facial recognition and illegally log into someone else’s account, exposing both property and privacy. At that point, it’s no longer at the level of a prank.

If I had to use an indie game as an analogy, this technology is a lot like Undertale—it seems light and clever on the surface, but can actually punch straight through the player’s psychological expectations. Creativity itself is not the original sin; the problem is that once it falls into the hands of scammers, the feedback loop is too short, and victims have almost no time to react. Compared with the “performative” feel brought by second-level processing in the past, millisecond-level face-swapping is much closer to real-time interaction. That means the old rule of thumb—“seeing is believing”—is rapidly losing its value.

My advice is simple, and more effective than installing some supposedly advanced software. First, try to avoid posting clear, unobstructed, front-facing photos and videos of your face on public platforms. Second, be cautious about granting permissions to unfamiliar apps and platforms, and don’t proactively hand your materials over to the training pool. Third, whenever a video call involves money transfers, borrowing money, or verification codes, no matter how much the other person looks like who they claim to be, you must verify through another channel—for example, by calling the original number directly, or by asking something that only the two of you would know.

From the trend, this probably isn’t going to reverse. It will only keep getting cheaper, faster, and more convincing. For ordinary people, the most practical “upgrade” is not learning how to spot flaws frame by frame, but building new verification habits. To put it bluntly, preventing fraud in the future may be less like fighting a Boss by relying on reflexes, and more like playing a puzzle game—you need one extra step of confirmation, one more layer of verification. It sounds troublesome, but compared with having your account stolen or your savings scammed away, this is one step you really cannot skip.