Two weeks ago, Will Smith posted a video of him performing “tour” on his Instagram. Since then, some sharp-eyed observers have noticed that the crowd appears to be made up of creepy AI horror people: lifting up banners covered in smooth shiny faces, messed up text, making them weird Sims– Like movement. Smith appears to be in some sort of Hollywood Shadow-banned longtail following a slap, but he tries to shrug moody sci-fi actor Will and joking mid-London, in order to revive his goofy dance with Carlton in the 1990s.
So we cannot shake up the feeling that we are witnessing advanced reputation management here. I spoke with KYT, the head of creative research at Kaiberai, as I was really created by AI to look into the crowds that included people who were grateful for the help they helped them endure cancer.
VICE: Just a little KYT! Is it video AI?
KYT: I think what we're looking at here is what I call a “real sandwich.” Some clips look real, but what's sandwiched between them are shots of these fake crowds. I think the root of the cloud shot – what you call “start keyframes” in AI is basically the image you use to kick off video generation – is the real thing. So I saw online people saying, “Oh, the signs are fake, blah, blah, blah.” I delved into it. It's a Euro Tour. I think the signs are Authentic. I think the actual starting image is Authentic. Now the clips they put in have probably cut them down to some degree, so they descend on a kind of mash. But I think about 60% of the videos are probably real and maybe 40% are generated. Why do you think it's a fake, what would help me tell you?
Certainly, what does that mean by being trimmed more and more?
The meaning I mean “trimmed” is that one of the existing limitations of these models now is the duration of the clips they can make. Usually, most of these models cannot make the clip longer than 10 seconds. So, when I think about this from the editor's point of view, what I imagined was that the editor had a concert footage of Will and needed more footage from that concert. I think they effectively put these shots in the editing timeline, realising they needed more, went to video models and took pictures of the crowds from other shows on this Euro tour. It looks almost daytime, as it changes dramatically, like some shots. It's really strange.
right.
The details are how I know it will be trimmed and descended onto the mash. So, when you look at the first part of the shot, the one of the crowd, there are some obvious signs that yes, it's AI. The timing and movements are a little unnatural. The timing of how their bodies move feels a little unnatural. They are very short shots and you can say that it's like they put a shot in the editor and when it started descending on the mash they cut it out.
Will Smith is consistently visible to himself on the video.
Looking at Will Smith, I think that's true. Will Smith is interesting as he has been used as a test subject for many AI video research. Will Smith eating spaghetti is like this long benchmark for video models. But, I think those parts are very realistic.
I think it's very bold for the editor to decide to use AI specifically for cloud shots. Perhaps one of the most difficult things is with regular 3D animations, especially in AI, as there are so many details. You've not only seen us already struggling with video models, but you're also trying to explain that everyone in the crowd has a phone, hat, shirt with a print, or sign. so. In most cases, models are not enough to handle that many of their characters and their complicated scenes, so if you want to create something incredible, we don't recommend using AI to fake such shots.
“There's nothing I hate here, but he's like a stubborn arc right now in public. Was that intentional?”
How can he expect people to not notice? Any star would be seen often, and I would say he… I don't really think of him as a music star, but these Europeans might do that at these concerts. Why does he do that?
He is a fresh prince. So I think it's very exotic to bring to Europe. If you read other people's analysis and see what the proportions of videos are not real, I think this is like an algorithmically driven decision. It feels like they padded out the runtime and created fake shots to hit Instagram's reel and the one-minute sweet spot that Tiktok demands in priority. And it's not really a new idea. Rats at Coachella 2023 Photoshop her crowd for an Instagram recap and she caught doing it [Latto denies photoshopping the images]. The fact that in the age where we exist now is more important than what actually happened, realizing and packaging the moment is really symbolic.
But in contrast to this, my second take – and I don't know what intentionality is here, but I know who Will Smith's media team is [Kyt believes it includes Jas Davis, a content creator known for his use of AI-generated imagery] Also, Smith is just a person, and there's nothing I hate here, but for now he's become like a fragment of the public… could it be intentional? Not rage bait, Argo– Feeding on a variety of things. The video appears to have been sitting for two weeks before people noticed. They were intentionally trying to do it, or they weren't very good at it, it was a sloppy edit.
I saw that after seeing spaghetti before, he reacted to it and made his own by eating actual spaghetti. So Will Smith knows that he knows AI.
And his team is very aware. So, his main content guy is someone who was traveling with Skrillex, like 15 years ago, ten years ago, and is like a content creator who is an assistant to Skrillex, and is a very media artist. My guess is that Will's team probably has important ears on the ground, embrace some of this new technology and knows that he can tie him into this bigger AI conversation. In the strange and messed up way, it's a kind of badge of honor that has become a benchmark for some of this AI test, like Will Smith eating spaghetti. And there's something almost appealing to using AI footage by retrieving it in your show if it doesn't come across anything so creepy and uncomfortable, because there's no better word.
Enjoy being able to identify this content as much as possible. One day you'll really miss these artifacts and this garbled text and these inconsistent movements. Because many directions of this technology are that my best advice for the future world is to live comfortably in a kind of pseudo-reality. It becomes increasingly difficult to detect things.
Follow Nick Thompson on Instagram @niche_t_
