(AFP) — OpenAI, the organization behind ChatGPT and the image generator DALL-E, has revealed ongoing testing of a text-to-video model named Sora.
This platform aims to enable users to create realistic videos with simple prompts. OpenAI, backed by Microsoft, showcased the capabilities of Sora in a blog post, mentioning its ability to generate videos up to a minute in length while maintaining visual quality based on user prompts.
Sora not only creates videos from textual prompts but can also generate videos from existing still images. According to OpenAI CEO Sam Altman, the platform is currently in a testing phase, with limited access granted to a select group of creators. Altman encouraged users to suggest prompts, sharing convincing results on the platform shortly afterward.
Some examples of Sora’s output include a video featuring two golden retrievers podcasting on a mountain and another showcasing a creature described as “half duck half dragon” flying through a beautiful sunset with a hamster in adventure gear on its back. However, OpenAI acknowledged that the current model has weaknesses, such as occasional confusion between left and right or difficulties maintaining visual continuity throughout a video.
Safety is a top priority for OpenAI, and the company emphasized that Sora would undergo adversarial testing, also known as red-teaming. This involves dedicated users attempting to make the platform malfunction, generate inappropriate content, or exhibit unexpected behavior. OpenAI plans to engage with policymakers, educators, and artists globally to address concerns and identify positive use cases for this new technology.
It’s worth noting that other tech giants, including Meta, Google, and Runway AI, are also actively developing text-to-video AI technology, with similar samples of their work being released.