Apple researchers have created Keyframer, a generative AI test app that allows users to describe an image and how it should animate.
Animation requires a more complex set of parameters, including scene duration and coordination of object movement, which are not easily captured in a single task description, so alternative means, including command clarification, may be required.
These goals were achieved in the Keyframer system – it creates animated illustrations from static two-dimensional pictures. Commands are processed by the GPT-4 neural network; An SVG (Scalable Vector Graphic) format image is used as the source, and the output is CSS code that animates the picture. The user can refine this code by manually editing it or by specifying an additional text command.
While the research paper—a 31-page, 16,000-word document—has been published, Keyframer itself has not been released and is purely an internal testing application.
However, its existence supports claims that Apple is actively testing generative artificial intelligence. Rumor has it that Apple will unveil significant AI-related improvements to iOS and Siri at WWDC 2024.
