OpenAI launches its anticipated AI video generator, Sora, on Sora.com today, but with limited availability.
The service will be available in the US and many other countries, but notably absent from most EU countries and the UK. While OpenAI CEO Sam Altman hasn't explained the regional restrictions, he indicated the wider rollout might take some time.
Update: On X, Altman attributed Sora's delayed EU launch to AI regulations in the region, though he didn't specify the exact reason. He indicated that due to these regulations, future OpenAI models might launch later in the EU—or not at all.
ChatGPT Plus and Pro subscribers can access Sora through their existing subscriptions at no extra charge, even though it's designed as a standalone product.
ChatGPT Plus subscribers are limited to 50 generations per month at 720p resolution, with each video capped at five seconds.
Pro subscribers get a better deal: unlimited slow-speed generations, 500 fast generations, and access to higher resolutions. These videos can run up to 20 seconds—though when OpenAI first announced Sora, they had mentioned videos could be up to one minute long. Only Pro subscribers can use Sora to animate uploaded photos that contain people.
The version launching today is what OpenAI calls a "turbo" variant—a streamlined version of the Sora model introduced in February, optimized for cost and speed.
The system lets users create videos in several ways: by typing text descriptions, turning still images into videos, or modifying existing videos.
It also includes features for remixing video content. The "Re-cut" feature lets users trim and extend individual scenes. Users can also seamlessly merge videos ("Blend") and create endless loops ("Loop") from their content.
Sora comes with several style presets, including a "Cardboard & Papercraft" option that transforms scenes to look like they're made of cardboard and paper, and a "Film Noir" look.
First review reveals impressive visuals, but physics still needs work
Tech reviewer Marques Brownlee (MKBHD) spent a week putting Sora through its paces and found both strengths and weaknesses. The AI shines at creating beautiful abstract scenes and cartoons, but stumbles when it comes to following basic laws of physics.
Objects in Sora's videos tend to appear and vanish at random or move in ways that defy physics, Brownlee found. This shows a key limitation of current AI: while it can match patterns to create convincing content, it doesn't actually understand how the physical world works—something OpenAI hopes to improve with future Sora research.
The physics problems show up most clearly when Sora tries to animate people and animals, with legs sometimes jumping into impossible positions. But it's not all bad news—Brownlee noted that the AI handles effects like flowing water and fire with remarkable accuracy.
Generation times vary by resolution: 480p videos take less than 20 seconds to create, while complex 1080p videos might need several minutes to complete.
According to Brownlee, Sora excels at creating abstract content, text animations, and cartoons. The cartoon style works especially well because any physical oddities can simply look like artistic choices.
While OpenAI has built in safety measures like watermarks, Brownlee worries about how the technology could be misused to create misleading content. He sees Sora as just the beginning of a future where AI-generated videos will be everywhere.
