What This Futuristic Olympic Video Says About the State of Generative AI

Because each shot requires a new set of prompts, it’s also difficult to create a sense of continuity in a video. The color, angle of the sun, and shapes of buildings are hard for a video generation model to keep consistent. The video also lacks close-ups of people, which Kahn says AI models still struggle with.

“These technologies are always better for large-scale things than for really nuanced human interaction,” he says. For that reason, Kahn thinks early film applications of generative video might be for wide shots of landscapes or crowds.

Alex Mashrabov, an AI video expert who left his role as director of generative AI at Snap last year to start a new AI video company called Higgsfield AI, agrees with the current failures and shortcomings of AI video. He also points out that good, dialogue-heavy content is difficult to produce with AI, as it often relies on subtle facial expressions and body language.

Some content creators are hesitant to use generative video simply because of the amount of time required to repeatedly prompt the models to get the end result right.