Natural language processing

OpenAI’s video AI Sora reportedly gets an upgrade to produce longer, higher quality clips faster

OpenAI’s video AI Sora reportedly gets an upgrade to produce longer, higher quality clips faster



summary
Summary

OpenAI is working on an upgraded version of its video AI Sora, which it unveiled in February, The Information reports, citing a source close to the company’s management team. The new version aims to produce longer, higher quality video clips faster than the original demos.

The video model presented in February was apparently not ready when first announced. Generations reportedly took over ten minutes, and feedback from filmmakers was sometimes critical.

Filmmaker Patrick Cederberg had to generate hundreds of clips before getting a usable one, as the model struggled to maintain consistent styles, objects, and characters throughout videos. Physics errors also occurred, according to The Information.

Such problems are common in other video models; the improved Sora version aims to address many of these issues. To improve Sora, OpenAI is collecting millions of hours of high-resolution video footage as training data, covering different styles and subjects to reduce bias, according to the source.

Ad

Sora gets competition

OpenAI told CNBC in early September that Sora was still in research mode due to ongoing discussions with policymakers. OpenAI CTO Mira Murati said in mid-March that Sora would be released “definitely this year,” though generation costs needed to be reduced first, as the system remains “much, much more expensive” than current AI systems.

Since Sora’s February launch, the video AI market has developed rapidly. In China alone, four new systems – KLING, Vidu, Jimeng AI and MiniMax – have been introduced. Runway ML has also presented two new AI models and recently announced a partnership with Lionsgate.

OpenAI's video AI Sora reportedly gets an upgrade to produce longer, higher quality clips faster

Source link