OpenAI has launched its latest artificial intelligence video generator, Sora, but has imposed restrictions on the depiction of people due to concerns about misuse. Users subscribed to the premium version of ChatGPT can now create AI-generated videos based on textual prompts, showcasing creative clips like sumo-wrestling bears and a cat sipping coffee. However, only a limited group of testers has the ability to generate videos featuring human subjects. This cautious approach aims to combat issues surrounding the misrepresentation of individuals and the proliferation of deepfakes, as articulated by OpenAI in a recent blog post.
Monitoring and Content Restrictions
The introduction of Sora comes with significant monitoring policies. OpenAI has prohibited content that includes nudity and is prioritizing the prevention of harmful applications, such as child sexual abuse material and deepfake pornography. These measures highlight the company’s commitment to ensuring the responsible use of AI-generated content in sensitive scenarios.
Key takeaways
- OpenAI released Sora, its text-to-video generator, to premium ChatGPT subscribers, but restricted the depiction of people.
- Only a limited group of testers can generate videos featuring human subjects, a deliberate limit aimed at misrepresentation and deepfakes.
- Nudity is prohibited outright, with particular emphasis on preventing child sexual abuse material and deepfake pornography.
- Demand was heavy enough that OpenAI temporarily disabled new Sora account creation.
- OpenAI has not disclosed which images and videos were used to train the model.
High Demand and Account Creation Pause
Since its release, Sora has seen overwhelming interest, prompting OpenAI to temporarily halt new account creation to manage the demand effectively. The company reported, “We’re currently experiencing heavy traffic and have temporarily disabled Sora account creation” on its website.
Background and Future Engagement
Sora was introduced earlier this year, with OpenAI aiming to consult with artists and policymakers before rolling out the tool to a wider audience. There is still uncertainty regarding the specific imagery and video sources utilized to train Sora, as OpenAI has not disclosed this information. The anticipation surrounding Sora underscores the growing role of AI in content creation and the importance of responsible practices in deploying such technologies.
Frequently asked questions
Why restrict videos of people specifically?
Realistic synthetic video of real individuals is the most direct route to harm — deepfakes, impersonation and non-consensual imagery. Limiting human depiction to a small tester group buys time to see how the safeguards hold.
What can subscribers actually make?
Videos generated from text prompts, with the examples cited running to sumo-wrestling bears and a cat drinking coffee — creative and clearly synthetic subjects rather than realistic depictions of people.
What remains unresolved?
Training data. OpenAI has not said what imagery and video the model learned from, which is the same disclosure question facing every major generative video system.
