Is it Real or AI? New Will Smith Eating Spaghetti Sora Video

Source: Will Smith eating spaghetti (Then vs Now)

As you may have noticed, the video below does not have the OpenAI watermark, which suggests that it is a fake AI video. But, surprisingly, it is actually Will Smith himself performing in the video.

Yes, you heard it right, “The Pursuit of Happyness” star himself went on camera and made a big show of it. He also pretended to look shocked and added the caption “This is getting out of hand!”, which fooled quite a few netizens. 

Many even watched the video 5 times before they realized what was actually happening. For instance, one user commented saying, the most creepy thing is that it’s hard to even tell whether it’s a real performance or AI-generated video. 

Even if the video is not AI-generated, it certainly has forced many to start to think about how confused AI would be when it learned this training in the future. 

Can Sora really do it?

Since Sora AI was announced, netizens have been @Altman OpenAI every day asking for a remake of the noodle-eating video. But, due to concerns about the safety of generating real human images, OpenAI has never responded to these requests. In the officially released Sora videos, the closest comparison is “a man eating a hamburger”. 

Source: Sora's video of a man eating a burger. Can you tell it's not real?

As you can see in this video, the bitten hamburger correctly shows bite marks demonstrating Sora’s understanding of the interaction between people and food. It’s unlikely that we will be able to see a real Sora-generated video of Will Smith eating noodles publicly. 

OpenAI’s red team testers are using Sora to find risks or potential harms in critical domains, akin to the approach taken with DALL-E 3, to mitigate the system’s ability to generate realistic human images as much as possible. It’s also worth noting that participating in the red team test is one of the only two ways to get early access to Sora. 

But, the test for participating in the red team test is also relatively high. To get access, you need to be an expert in a certain field. Aside from this, the only way to play with Sora is to become an official OpenAI employee. 

How do they make AI videos? Similar to Will Smith AI 

After the recent announcement of OpenAI's Sora video synthesis model, many have noted the dramatic jump in AI video quality over the past year compared to the infamous spaghetti video. Smith's new video plays on that comparison by showing the actual actor comically eating spaghetti and claiming that it is AI-generated.

We have not yet seen a model with the capability of Sora attempting to create a new Will-Smith-eating-spaghetti AI video, but the result would likely be far better than what we saw last year, even if it contained obvious glitches. 

Given how things are making progress, we wouldn't be surprised if by 2025, video synthesis AI models can replicate the parody video created by Smith himself.

It's worth mentioning for history's sake that despite the comparison, the video of Will Smith eating spaghetti did not represent the state of the art in text-to-video synthesis at the time of its creation in March 2023 (that title would likely apply to Runway's Gen-2, which was then in closed testing).

However, the spaghetti video was reasonably advanced for AI models at the time, having used the ModelScope AI model.

In a Nutshell 

Although the viral “Will Smith eating spaghetti” video is a clever fake, it highlights the potential for realistic AI-generated content. While Sora AI isn’t currently generating real human images due to safety concerns, its ability to understand interactions like biting a hamburger shows its progress.