When OpenAI launched its AI generated video platform earlier this week, it did so while admitting that the system wasn’t yet very good at complex physical movements. Now a viral video of a gymnast that distorts limbs and even appears to lose her head as she twists through the air illustrates Sora’s known issue with a mix of horror and hilarity.
The video, which now has more than 1.7 million views on X, was created and shared by Deedy Das, a venture capitalist. “As cool as the new Sora is, gymnastics is still very much the Turing test for AI video,” the caption reads. Das previously shared an equally horrifying AI gymnastics video over the summer created by the AI platform Luma.
As cool as the new Sora is, gymnastics is still very much the Turing test for AI video.1/4 pic.twitter.com/X78dNzusNUDecember 10, 2024
Ars Technica calls the poor understanding of physics the “jabberwockies,” but quotes Das as calling the Sora video an improvement because it lacked the teleportation and random outfit changes of his previous videos generated by Luma.
When OpenAI launched Sora, it did so sharing known limitations, saying that it “generates unrealistic physics and struggles with complex actions over long durations.” The gymnastics video is a key example of how far the generator needs to grow to create videos that follow the laws of physics.
As a photographer and mother of a daughter who has participated in gymnastics in the past, I watched the video with a mix of both horror and fascination. At times, the athlete's head seems to reattach to a different part of her body and arms and legs appear interchangeable.
When I wasn’t watching the gymnast’s arms and legs morph from a human into what can only be described as an animated Mandala, I recognized the photographer in the background of the scene, who has four hands: two resting on a bench and two on a camera. As the gymnastic makes its way across the mat, an identical twin four-handed photographer watches from a completely different spot.
The photographer also wasn't looking at the athlete’s physics-defying movements. The AI generated photographer instead stares straight ahead looking rather bored, not following the performance with eyes or lens. So not only is AI not advanced enough to replace a videographer, it can’t even imitate one correctly.
Admittedly, I have thought at times that it would be helpful to have an extra set of hands – but that’s not at all what I was envisioning.
You may also like
While you're laughing at generative AI fails, be sure to read up on how to tell if an image was AI generated – because some of them aren't so obvious. Or, take a real video with one of the best cameras for video.
