OpenAI Just Released Its AI Video Generator, Sora

After months of anticipation, OpenAI has released Sora, its first-ever AI model designed for text-to-video and image-to-video generation. In a live streamed video presentation, OpenAI CEO Sam Altman announced the model’s launch, available now at Sora.com. The release of Sora is arguably OpenAI’s biggest launch of 2024—one the company’s been teasing since first revealing the model in February. Only a select number of customers have had access to Sora since then, such as Toys “R” Us, which debuted the first Sora-created ad in June. In an early review, technology influencer Marques Brownlee called Sora “horrifying and inspiring at the same time.” Sora users will be able to generate video in different resolutions, from 480p (SD) to 1,080p (HD), with higher resolutions taking more time to generate. The size dimensions, length, and speed of the video can also be customized. In addition, users will be able to see other people’s AI-video creations and then remix or alter them. In an example, Brownlee successfully altered a video of a house on a cliff to add a golf course to the background. Users will also be able to upload images and ask Sora to turn them into videos. Brownlee says he found the most success by generating images with OpenAI’s Dall-E, and then uploading them to Sora. Brownlee says there are still some major areas where Sora isn’t ready yet. The model struggles with object permanence—with objects often blipping into and out of existence—and it hasn’t quite figured out how to flawlessly recreate physics. The videos also don’t currently include sound of any kind. As for how this first version of Sora can best be used commercially, Brownlee suggests using the model to create abstract videos and title designs. Sora can generate incredibly detailed textures and complex patterns, so it’s especially good at generating the kind of eye-catching abstracts that can often be found on modern websites. Plus, Brownlee says Sora can be quite accurate at creating titles or logos when given specific words to recreate. Altman says that Sora will be available for users in the United States later today, but access in the U.K. and most of Europe will take some time. According to Sora’s product page, ChatGPT Plus subscribers, who pay $20 per month, will be able to generate 50 videos per month with a 720p resolution and a maximum length of five seconds. ChatGPT Pro subscribers, who pay $200 per month, will be able to generate unlimited videos, and make videos with resolutions of 1,080p and a maximum length of 20 seconds. Pro subscribers will also be able to download videos without a watermark.