Google launched its scarily-good new AI video model Veo 3 last week — and users wasted no time putting it to the ultimate benchmark: the “Will Smith Eating Spaghetti” test.
The Will Smith Eating Spaghetti test has become an unofficial rite of passage for generative video models and a baseline for gauging how powerful AI models really are.
Just got access to Veo 3 and the first thing I did was try the Will Smith spaghetti test. SOUND ON pic.twitter.com/y0CiZwNxgM
— Javi Lopez ⛩️ (@javilopen) May 22, 2025
The informal benchmark and internet meme originated about two years ago, when early attempts at AI-generated video struggled with realism — especially with coherent human faces and complex actions like eating.
In March 2023, distorted and nightmarish AI-generated videos of Will Smith awkwardly eating spaghetti — featuring strange facial movements and uncanny visuals — went viral online.
AI videos in 2023 vs. 2025
We all laughed at AI Will Smith eating spaghetti.
Now, look at the quality we have today.
It’s not perfect, but if AI can achieve this in less than a year, imagine what will be possible in the next few years or even months…
Google’s Veo 3 is… pic.twitter.com/fFCxBJHCIZ
— Endrit (@EndritRestelica) May 22, 2025
Since then, Will Smith eating spaghetti has become a humorous litmus test for generative video technology — even the actor parodied the trend in an Instagram post last year. Testing whether a new video generator can realistically render Smith slurping down a bowl of noodles quickly became a viral shorthand for how far AI still had to go in producing lifelike human actions, coherent facial expressions, and messy, nuanced tasks like eating.
When Google unveiled its advanced Veo 3 last Tuesday, users immediately began applying the test and it was clear that generative video technology had dramatically improved since Smith’s first viral moment eating spaghetti in 2023.
in the future, people emerging from comas will gauge AI progress through the will smith spaghetti test pic.twitter.com/E8iZRcaHXR
— Shawn Kanungo (@ShawnKanungo) May 22, 2025
One particularly lifelike clip made using Veo 3, shared by AI content creator Javi Lopez, shows a recognizable Smith eating and slurping noodles. Although Google’s model includes AI-generated sound, the actor bizarrely makes crunching and squelching noises while chewing the spaghetti in the video — a bizarre detail that led YouTuber Marques Brownlee to comment “I don’t feel so good.”
The inaccurate crunching sound is likely due to a glitch in Veo 3’s experimental ability to apply sound effects to video. According to Ars Technica, the strange sound effects are probably a result of training data that included many clips of chewing mouths paired with exaggerated crunching sounds.
Veo 3 is so good that it even does a decent Will Smith being eaten BY the spaghetti. pic.twitter.com/ye2qJbqJxs
— Jason Rink (@TheJasonRink) May 24, 2025
However, users were mostly impressed by Veo 3’s capabilities at making Smith eat spaghetti — and soon decided to flip the script, challenging the model to generate clips of spaghetti eating the actor instead.
Veo 3 was announced at Google I/O 2025 last week. Imagen 4, the company’s AI image generator, was launched too.