Two weeks ago, Will Smith posted a video of him performing “on tour” on his Instagram. Some keen-eyed observers have since noticed that the crowd appears to be made up of uncanny AI horror-people: smooth glossy faces, holding up banners covered in fucked-up text, making weird Sims-like movements. Smith, who still seems to be in the long tail of some kind of Hollywood shadow ban following The Slap, has been flailing a bit of late, trying to shrug off Moody Sci-Fi Actor Will in order to resurrect Goofy Dancing With Carlton in the 1990s Will, playing to bemused crowds of grimacing locals in the middle of the street in London, and releasing jarring singles about being a horny 56-year-old.
As such, I can’t shake the feeling that we are witnessing some advanced reputational management here. To find out if the crowd—which includes someone waving a banner, thanking Will for helping them survive cancer—really were created by AI, I spoke to Kyt, head of creative research at KaiberAI.
VICE: Hey Kyt! Is the video AI?
Kyt: I think what we’re looking at here is what I would call “a reality sandwich.” It looks like some of the clips are real, and then sandwiched in between are these fake crowd shots. I think the root of the crowd shots—what we would call in AI the “starting key frame,” basically the image you use to kick off the video generation—is also real. So I saw some stuff online of people saying, ‘Oh, the signs are fake and blah, blah, blah.’ I did some digging. It’s a Euro tour. I think the signs are real. I think the actual starting images are real. Now, the clips that they put in, they probably trimmed them down to some degree, so they kind of descend into mush. But I would say about 60 percent of the video is probably real, and maybe 40 percent is generated. Would it be helpful for me to tell you why or what makes me think that it’s fake?
Sure—also, what do you mean by trimmed down so much that it’s mush?
What I mean by “trimmed” is that one of the existing limitations of these models right now is the duration of the clips they can make. Typically, most of these models cannot make clips longer than ten seconds. So if you think about this from an editor’s perspective, what I imagine happened was the editor had some concert footage of Will performing, and they needed more of that concert footage. I think they effectively put these shots into an editing timeline, realized they needed more, then went to a video model and took photos of the crowd from other shows on this Euro tour. The reason I think that is because the lighting varies dramatically—like, in a couple of the shots, it almost looks like it’s daytime. It’s really weird.
Right.
The way I know that it’s trimmed and descending into mush is the details. So the beginning part of the shot, when you see the crowd stuff—yes, there are some obvious signs that it’s AI. The timing, the motion is a little unnatural. The timing of the way that their bodies move feels a little bit unnatural. They’re very, very short shots, and you can tell that they sort of put the shot into the editor, and when it started to descend into mush, they clipped it.
I notice Will Smith looks consistently himself in the video.
I think when you see Will Smith, that’s real. It’s funny because Will Smith has been used as a test subject in a lot of AI video research. Will Smith eating spaghetti is like this long benchmark in video models. But no, I think those parts are very much real.
I think it’s incredibly bold for an editor to decide to use AI specifically for crowd shots. It’s probably one of the hardest things to do, both in regular 3D animation, but especially in AI, because there’s just so many details. You’re trying to account for not just one human moving naturally, which we’ve already seen video models struggle with, but a crowd of them, and then every person in the crowd has a phone or a hat or a shirt with a print on it, or a sign—it’s so much. I would not recommend using AI for faking shots like that if you want to make something believable, because the models aren’t yet good enough to handle that many figures or that complex of a scene, most of the time.
“I mean no hate here, but he’s sort of on like a cringe arc in the public right now… I wonder if it was intentional.”
How could he expect people not to notice? Any star is going to be watched closely, and I guess he’s a… I don’t really think of him as a music star, but I guess these Europeans maybe do in these concerts. Why would he do it?
He’s the Fresh Prince. I mean, I think it’s a very exotic thing to bring to Europe. I think in reading other people’s analyses and looking at what percentage of the video is real versus what isn’t, that this was sort of an algorithmically driven decision. It feels like they made fake shots to pad out the runtime and hit that one-minute sweet spot that Instagram reels and TikTok prioritize and ask for. And that’s not really a new idea. Latto at Coachella 2023 photoshopped her crowds for her Instagram recaps, and she got caught doing it [Latto denies photoshopping the images]. It’s really emblematic of this idea that the era we exist in now, the perception and packaging of the moment matter more than what actually happened.
My second take, though, in contrast to this—and I don’t know what the intentionality is here—but knowing who Will Smith’s media team is [Kyt believes it includes Jas Davis, a content creator known for his use of AI-generated imagery] and also Will Smith just being somebody who’s sort of on, I mean no hate here, but he’s sort of on like a cringe arc in the public right now… I wonder if it was intentional. Like, not rage-baiting, but like, an algo-baiting kind of thing. The video apparently sat for two weeks before people noticed. Either they intentionally tried to do that, and they’re not very good at it, or it was sloppy editing.
I saw the spaghetti thing earlier, and then saw that he responded to it and made his own thing, eating actual spaghetti. So we know Will Smith is aware of AI.
And his team is very much aware. So his main content guy is somebody that I used to know or worked with, like ten, 15 years ago, who used to travel with Skrillex, and was sort of Skrillex’s assistant, content creator, and is very much a media artist. My guess is that Will’s team probably has a significant ear to the ground and is both embracing some of this new technology and knows that they can tie him back to this larger AI conversation. In some weird, messed-up way, it’s kind of a badge of honor that Will Smith has sort of inadvertently become a benchmark for some of this AI testing, like the Will Smith eating spaghetti thing. And there is something almost charming about taking it back and using AI footage in your shows, if it just didn’t come across so uncanny and off-putting, for lack of a better word.
Savor being able to identify this content while you can; one day, you will really miss these artifacts and this garbled text and these inconsistent movements, because the direction a lot of this technology is going, like, really, my best advice for the future world is to get comfortable living in a sort of pseudo reality. Detecting things will get harder and harder.
Follow Nick Thompson on Instagram @niche_t_
The post Did Will Smith Really Create AI Fans to Make Himself Look More Popular? We Asked an Expert appeared first on VICE.
