Sora’s Capabilities and Limitations
OpenAI launched its video generation program, Sora, for customers on Monday. The tool converts written prompts into videos up to 20 seconds long. OpenAI initially released Sora as a beta in February and now offers it as a standalone product.
CEO Sam Altman emphasized the significance of video in AI advancements during a live-streamed announcement. “We don’t want the world to just be text. AI systems should also interact with visuals,” Altman said.
Sora uses a deep understanding of language to create detailed and dynamic videos. It can generate complex scenes with multiple characters, camera angles, and specific motions. Examples include animated monsters, realistic scenes of people, and cinematic trailers. However, the program has limitations, such as unrealistic physics and struggles with long, complex actions.
Focus on Responsible Use
OpenAI aims to lead in shaping AI video generation’s culture and rules. “We’re introducing this technology to help society explore its potential and develop safeguards for responsible use,” the company said in a blog post.
The company acknowledged concerns about AI abuse, including deepfakes. OpenAI restricts uploads of real people but plans to ease these limits as its mitigation tools improve. Its top priority is preventing harmful content, such as child abuse material and explicit deepfakes, by employing filters, detection tools, and reporting mechanisms.
To promote transparency, Sora embeds metadata and watermarks in all videos by default, allowing users to identify AI-generated content.