Tuesday, Google Launted VEO 3A new AI video synthesis model that can’t do anything is a big AI video generator has not been able to do it before: Create sync audio track. While from 2022 to 2024, we saw preliminary steps in the AI video generation, each video was quiet and generally very short in the period. Now you can hear sounds, dialogue and sound effects in eight seconds high definition video clips.
Immediately after the new launch, people started asking the most obvious bench marking question: How good is the Oscar -winning actor in a fictitious food to eat spaghetti?
First, a short recovery. In the AI video, Spagiti Benchmark tracked its origin by March 2023, when We covered first The initial example of the horrific AI-generation video using the open source video synthesis model. Spaghetti example later became quite famous about Smith What heyed to him About a year later in February 2024.
Here the original viral video looks like:
One thing that people forget is that at that time, the example of Smith was not the best AI video generator there. Runway Has already achieved high results (though it was not yet publicly accessible). But the result of the models was ridiculous and so strange that the early faulty example of video recipe could be stood in people’s memories, easy to compare the future as soon as AI models developed.
AIAP developer Javi Lopez arrived with the Smith test earlier this week to rescue curious spaghetti fans earlier this week and Posting results X on. But as you see when you see, the sound track has a curious standard: the wrong Smith is crushing on spaghetti.
On X, Javi Lopez received “Will Smith ate Spagite” in Google VEO 3AI video generator and received the result.
This is a defect in the experimental ability of VEO 3 to apply sound effects on the video, it is likely that the training data used to create Google’s AI model contains many examples of mouth chewing mouth with sound effects. The Generative AI Model Pattern is the forecast machines, and they need to show enough examples of a variety of media so that the new output can be created. If a concept is more represented or represented in training data, you will see the results of an extraordinary generation, such as Jaber Vocyes.