OpenAI, a company supported by Microsoft, is currently developing software designed to create one-minute videos in response to textual prompts. The company, created by former Google executive Sam Altman, showcased the tool Thursday at its X event, demonstrating photorealistic details of fantasy scenes from prompts. It’s a significant step beyond previous text-to-video models, typically creating shorter clips or filling in missing frames of existing video footage.
During its demonstrations, the OpenAI team also highlighted the system’s ability to create objects that appear from nowhere and interact with one another in realistic ways. For example, a basketball in the middle of a net that’s about to explode shoots straight through it like a ghost. In other demos, the system also caused dogs to run off into the distance and a bird to fly past the camera.
What’s most impressive is that Sora can produce these minute-long videos while maintaining visual quality and adherence to the prompt, the company said. Other text-to-video systems can only create up to 20 seconds at most, according to TechTarget. The system is currently undergoing “red teaming,” which means it’s being used by experts who identify flaws in the AI model and by visual artists, designers, and filmmakers to gain feedback on the tool before it’s made available for general use.
Sora can also create a video from a still image or even extend a video by filling in missing frames, a technique called inpainting. This can be especially helpful for people who work in film and video and may need to edit or retouch videos. The company says it has plans to introduce other features in the future that will allow Sora to do even more creative things, including bringing back dead animals and morphing characters.
The team at OpenAI is aware of the potential implications of its new model, and it’s taking steps to protect users from misinformation and other harms that the technology could produce. For example, it is working on tools that will make it easier to detect when Sora generated a video, and the company will also include metadata in such content that would help consumers distinguish between real-world footage and computer-generated videos.
The company is evaluating whether to make the tool available for commercial use. However, it’s still in its early stages, and many challenges are ahead before becoming widely accessible. The Federal Trade Commission has proposed rules to make it illegal for businesses and government agencies to use AI-generated impressions of a person without their permission. There have already been examples of fake news, hate speech, and other forms of misleading information being distributed via AI video. The company has yet to announce a release date for the public version of Sora. Still, it will be released to “an exclusive group of visual artists, designers, and filmmakers” to get feedback on how the tool can be most beneficial to them.
