A dystopian AI-generated video of gymnasts performing proves that text-to-video technology isn't all that great yet.
The humorous video has been going viral online and was created by Wright Bagwell using Luma AI's free AI video generator Dream Machine.
The video shows the AI gymnasts performing strange movements, their bodies often headless and splitting into starfish-like human limbs, appearing and disappearing periodically.
This new AI gymnast is awesome! pic.twitter.com/YJuU151C29
— Daniel Feldman 🟥 (@d_feldman) June 29, 2024
“If you look closely at the background you can tell it's an AI,” one joker commented under Bagwell's post.
Over at X (formerly Twitter), Tim Spalding speculated that one of the reasons the AI footage was so terrible was because the AI models weren't trained on enough videos of gymnasts. This hypothesis is probably correct.
Bagwell has also used Dream Machine to create a series of other AI-generated sports videos, including cycling, wrestling, figure skating and table tennis.
This terrible AI video is reminiscent of earlier AI imagery, such as that produced by OpenAI's DALL-E 2, which is also notoriously unstable.
The early images coming out of DALL-E have developed a sort of cult fanbase, with fans reveling in the imperfect nature of the AI content. Many were disappointed when OpenAI recently shut down DALL-E 2, even though some had latched on to its dreamy, shadowy aesthetic.
But just like DALL-E 2, which was vastly improved upon by Midjourney and DALL-E 3, Luma AI's Dream Machine will likely soon be replaced by a better model that's less quirky and more consistent.
AI video is still in its early stages, but it seems inevitable that these kinds of synthetic images will start to compete with real videos and trick people into believing they are real.
The implications of this are huge, with fears that deepfake videos could be used by fraudsters to trick people or make politicians take actions they did not take to get people to vote in a certain way.
The most talked about AI video generator is Sora from OpenAI, which was announced a few months ago. As of this writing, it still doesn't have a release date. Behind the scenes, there seem to be concerns about the technology and discussions about how Sora was trained, which OpenAI refuses to reveal details about.
