The company that made ChatGPT, OpenAI, has announced the release of Sora, a new video-making software. The AI company said Thursday that Sora can make realistic and imaginative scenes from text directions.
Sora is a text-to-video model that lets people make photorealistic movies up to one minute long by telling the tool what to do.
While Runway and Pika have made great strides with their text-to-video models, Meta and Google’s Lumiere are OpenAI’s main competitors in making videos with AI tools. Like Sora, Lumiere gives users text-to-video tools and lets them make videos from still images.
Other companies make similar AI tools, like Stability AI, which has a product called Stable Video Diffusion. Amazon has also created Create with Alexa, a model that makes short cartoon children’s shows based on prompts.
Read also: OpenAI CEO refutes update leaks, speculations
More Sora info
OpenAI’s DALL-E image-generation AI programme is similar to Sora, according to a Thursday announcement.
How it works: Sora returns an HD video clip after a user types a scene. Sora may create still-image-inspired video clips and extend or fill in missing frames. The model can “accurately interpret props and generate compelling characters that express vibrant emotions” and grasp how objects “exist in the physical world.”
According to the OpenAI blog article, it can create “complex scenes with multiple characters, specific types of motion, and accurate details of the subject and background.”
An aerial view of California during the gold rush, a Tokyo train footage, and others are Sora-generated samples connected to the announcement.
One sample shows a strangely moving floor in a museum movie, but OpenAI notes the model “may struggle with accurately simulating the physics of a complex scene and may not properly interpret certain instances of cause and effect.”
Sora is now only available to “red teamers” investigating the model for hazard and risk. Some visual artists, designers, and filmmakers can contribute input to OpenAI.
Addressing safety
OpenAI claims it will take many safety procedures before adding Sora to its products.
We are collaborating with red teamers, experts in misinformation, hateful content, and bias, to test the model. Company stated.
The company is developing technologies to detect deceptive content, including a classifier to identify Sora-generated videos and leveraging safety approaches for DALL·E 3 products.
In an OpenAI product, our text classifier will evaluate and reject text input prompts that violate our usage restrictions, such as severe violence, sexual content, hostile images, celebrity likeness, or IP.
Finally, OpenAI will consult politicians, educators, and artists worldwide to highlight concerns and positive use cases for the new technology.