Google AI creates lifelike Will Smith double eating virtual spaghetti

Google’s Veo 3 marks a significant leap in AI video generation by introducing synchronized audio capabilities, a feature previously absent from major AI video tools. This development enables AI-generated videos to include voices, dialog, and sound effects within eight-second HD clips, significantly advancing the realism of synthetic media. The evolution from the primitive, silent AI videos of 2022-2024 to today’s more sophisticated audiovisual creations demonstrates the accelerating development of generative AI technology.

The big picture: Google’s new Veo 3 AI model represents the first major AI video generator capable of creating synchronized audio tracks alongside video content.

• The model can generate coherent dialog, music, and sound effects, dramatically enhancing the realism of AI-generated videos.
• This capability marks a significant advancement from the silent, short-duration AI videos that dominated from 2022 to 2024.

The spaghetti benchmark: AI developers have informally adopted “Will Smith eating spaghetti” as a standard test for evaluating AI video generation capabilities.

• This benchmark originated in March 2023 when early, unsettling AI-generated videos using the ModelScope model went viral.
• The test became culturally significant enough that Will Smith himself parodied it in February 2024.

Key details: App developer Javi Lopez conducted the “Will Smith test” with Veo 3, revealing both impressive capabilities and notable glitches.

• The AI-generated Smith appears to be crunching on spaghetti rather than eating it normally.
• This audio glitch likely stems from Veo 3’s training data containing numerous examples of chewing sounds paired with crunching audio effects.

Why this matters: Despite its imperfections, Veo 3 demonstrates remarkable progress in generating realistic AI content that includes coherent dialog and appropriate audio.

• Many examples shared on social media platform X showcase videos with wildly realistic audiovisual synchronization.
• The technology’s rapid improvement since 2023 suggests AI video generators will become increasingly sophisticated and convincing.

The concerning part: Google’s model currently includes celebrity filters that limit some applications, but the technology’s advancement raises important questions about synthetic media.

• Without such filters, the technology could potentially create convincing videos of real people saying or doing things they never did.
• As these tools become more accessible and capable, the potential for misuse through deepfakes and misinformation grows correspondingly.