OpenAI’s new Sora 2 AI model is a total mess

OpenAI promised Sora 2 would be the "most powerful imagination engine ever built." What the world got instead was a tangle of copyright confusion, awkward likenesses, and recycled characters.

Within hours of its launch, the internet filled with videos made using the AI model. Instead of fresh creations, familiar faces took centre stage: Sonic the Hedgehog, Pikachu, Solid Snake. Even when users did not ask for existing material, the model seemed unable to resist. A prompt for a woman riding a dragon in a flower world came back looking suspiciously like The NeverEnding Story.

Some characters are blocked though. Disney refused to allow its catalogue, so no Mickey Mouse or Darth Vader. But others slipped through easily. Users churned out clips in the style of Rick and Morty or even full echoes of Netflix's Wednesday. That raised questions about what Sora 2 was trained on, since Netflix and others have denied giving OpenAI permission.

The chaos is not limited to fictional characters. OpenAI chief Sam Altman gave approval for his likeness to be used. Users immediately dropped him into bizarre settings: a shoplifting spree in a Target shop, a skibidi toilet, even a thief stealing art from Hayao Miyazaki while Miyazaki is running after him. While funny, it showed how quickly the tool can turn a real person into a digital prop.

Logos and brands are not safe either. Videos have pulled in NFL team marks and retailer storefronts without clearance. And although OpenAI says users can control or delete videos of their likeness, exported clips can still spread unchecked.

Sora 2 does generate videos with remarkable polish. But its early use makes it evident that rather than a clean showcase of imagination, the model has become a messy collage of borrowed worlds, brands, and faces.