Will Smith's test eating spaghetti goes increasingly viral

The 'Will Smith eating spaghetti test' is one of the most curious - and revealing - benchmarks of contemporary artificial intelligence. It originated in 2023, when a Reddit user published an AI-generated video in which actor Will Smith was eating spaghetti: the result was grotesque, with deformed faces, impossible movements and pasta that seemed to merge with his body. That visual horror became a symbol of the limits of automatic video generation.

Since then, the 'Will Smith eating spaghetti' prompt has become an informal test to measure the progress of text-to-video models: credibly reproducing a human face while performing a complex action - eating - requires coordination between anatomy, physics, facial expressions and temporal continuity. In practice, it is a maturity test for visual AI.

After the uproar caused by the video of Tom Cruse and Brad Pitt published on X that misled many people, the search for a method to discover the veracity of audio-videos on the net is becoming more and more pressing. Not least because in the meantime, AI-powered video generators are on the rise. Cruise and Pitt's fistfight, for instance, was made with a two-line prompt inserted in Seedance 2, the new video generation model of ByteDance, the Chinese company that owns TikTok.

The leap in quality and realism, however, came in 2025 with Google's Veo 3, which was able to produce much more realistic clips than in the past, albeit with minor shortcomings (e.g. unnatural chewing sounds). Today, the competition is global: models like OpenAI's Sora, Google's Veo and Chinese systems like Seedance or Kling are all proving that they can 'pass' the test, with results that are increasingly indistinguishable from the real thing.

The point is not pasta, but trust: if a trivial video becomes indistinguishable from a real one, the entire visual information ecosystem changes. The spaghetti meme has thus become a cultural indicator of how fast AI is learning to simulate the world.