The videos generated by Sora feature highly detailed scenes, complex camera motion, and multiple characters with vibrant emotions.
The AI system can quickly create videos up to a minute long that represent complex scenes with accurate details of the subject and background. However, OpenAI is taking steps to ensure the safety and accuracy of the generated content before making Sora publicly available. The company is working with experts to assess potential harms, including the adversarial testing of the AI model.
Sora is currently being released to a small group of developers for testing12.The AI model, Sora, is designed to respond to text prompts by generating videos that exhibit an emergent grasp of cinematic grammar and storytelling. It is important to note that Sora may have limitations, such as struggling with simulating the physics of a complex scene and understanding instances of cause and effect. Additionally, it might have difficulty with spatial details from certain prompts, such as confusing left and right or struggling with precise camera trajectories25.

OpenAI is testing Sora, its new AI video model, through a process called red-teaming, which involves adversarial testing to ensure that the model does not produce harmful or inappropriate content. The company is also granting access to a select group of “visual artists, designers, and filmmakers” to gather feedback on how to advance the model to be most helpful for creative professionals. This testing phase is aimed at ensuring that creative professionals can benefit from the technology and to assess potential harms, including the adversarial testing of the AI model. OpenAI has not shared a specific timeline for a widespread release of Sora and is being careful about its deployment to ensure safety and accuracy before making it publicly available1.

Sora, OpenAI’s new AI video model, is distinguished from other AI video generators by its ability to create videos that appear more realistic and less like a sequence of AI-generated images. Sora’s videos are described as feeling like real videos, featuring highly detailed scenes, complex camera motion, and characters with vibrant emotions, which sets it apart from competitor models that may produce videos resembling stop-motion of AI images1. Additionally, Sora can generate videos up to a minute long, maintaining coherence and avoiding “AI weirdness” in terms of physical impossibilities, which is a common issue in other text-to-video models5. The model’s understanding of language and the physical world enables it to accurately interpret prompts and generate compelling characters that express vibrant emotions5. Sora’s capabilities include an emergent grasp of cinematic grammar and storytelling, allowing it to create videos with a narrative thrust, multiple shot changes, and precise camera angles and timing4. However, Sora may have limitations in accurately simulating the physics of complex scenes, understanding cause and effect, and managing spatial details from certain prompts25. OpenAI is also taking steps to ensure the safety and accuracy of the content generated by Sora before making it publicly available, including adversarial testing and working with experts to assess potential harms2.