Okay, here’s a revised version of the article, incorporating verification of claims and corrections where necessary. I’ve focused on ensuring accuracy and updating information as of today, February 10, 2024.
—
## The “Spaghetti Test” for AI video Generation is Getting Harder to Pass
If you recall, the very first version of AI Will Smith eating spaghetti was made with ModelScope and demonstrated the early challenges of maintaining facial consistency in generated videos. By the following year, the video – and countless variations of it – had taken off as a meme, to the point that Smith himself poked fun at it. He later used generative AI in a TikTok video of his own. Here’s an example of the test using Veo 3.1 from last year.
This Tweet is currently unavailable. It might be loading or has been removed.
Among today’s major players in video generation, like Grok and OpenAI, passing the “spaghetti test” – a benchmark for video generation quality and consistency – has become much harder. These companies have implemented strict guardrails to prevent the generation of content using third-party likenesses and copyrighted material, notably in response to concerns and legal challenges from Hollywood regarding AI models trained on intellectual property.
Mashable attempted to recreate the test using OpenAI’s Sora and Google Gemini’s Veo 3.1, but both attempts were blocked due to copyright restrictions. As AI generators, particularly those based in the U.S., increasingly restrict the use of recognizable faces and copyrighted elements, the “spaghetti test” in its original form might potentially be becoming obsolete.
Topics
Artificial Intelligence