There’s this music video made with OpenAI’s Sora model, which is pretty cool but I think also showcases its limitations. It looks quite realistic, but for anything longer than a few seconds, the tech is mostly only good for a stylized fever dream / infinite zoom aesthetic.

It’s a better concept than the movie described in this post though, which seems to just be trying to pass off the use of chatgpt instead of writers for the script as some clever critique on the use of AI. You know, by unironically doing what they’re critiquing.

For actual coherent long-form AI-generated video, I think something fundamentally needs to change about the approach to training.