First text, then images, now OpenAI has a model for generating videos.
The developers of ChatGPT and DALL-E announced this on Thursday Sora,a text-to-video dissemination model. From today, Sora is available to red teams, i.e. experts who adversarially test the model for damage and threats. According to the announcement, it is also available to a select group of visual artists, designers and filmmakers “for feedback on how to improve the model to make it most useful to artistic professionals.”
The tweet may have been deleted
OpenAI has been developing generative AI tools at a breakneck pace since ChatGPT launched in November 2022. Since then, we have released GPT-4, voice and image prompts, the modern DALL-E 3 image model, all available via ChatGPT. The OpenAI API has also had an exponential impact on the AI industry, enabling companies and developers to create their own generative AI tools. Now OpenAI is taking another significant step towards enhancing AI capabilities through video generation.
There are other video generation models, but none can produce realistic, convoluted videos like the ones Sora claims to create. Meta has a tool for creating low video clips, and Google is working on its own text-to-video model, but it is still in the research phase.
With Sora, users can generate videos up to one minute long, featuring detailed scenes and multiple characters. The announcement includes clips of the SUV following a winding mountain road and “historical” footage of California during the gold rush era.
Now for the security fine print. In addition to model assembly, OpenAI says its authoring tools allow videos created by Sora to be marked as such, in line with C2PA guidelines. It also leverages existing DALL-E security methods that reject inappropriate or malicious text messages.
Finally, OpenAI says it will “engage policymakers, educators and artists around the world to understand their concerns and identify positive exploit cases for this modern technology.” The company believes that the best way to understand all the beneficial and harmful ways people will exploit Sora is “learning from real-world applications is a key part of building and releasing increasingly secure AI systems over time.”
