Apple has thrown its weight behind “Keyframer,” a newly revealed AI tool that breathes life into static images using your words. This prototype tool allows users to add animation to 2D images simply by describing their desired motion. Using GPT-4 3, Keyframer generates CSS animation code to animate an input Scalable Vector Graphic (SVG). To support user refinement of generated designs, Keyframer has multiple editor types for users to directly edit generated animations.“Additionally, users can iterate on their designs through sequential prompting and request variants from the LLM to ideate on new design directions. Through these features, Keyframer accommodates users exploring and adapting their design goals as they iteratively construct an animation through combined prompting and editing actions,” Apple researchers said in a paper shared on the AI tool.
The researchers also explained how the tool is different from image generating options like Dall-E. “While one-shot prompting interfaces are common in commercial text-to-image systems like Dall·E and Midjourney, we argue that animations require a more complex set of user considerations, such as timing and coordination, that are difficult to fully specify in a single prompt—thus, alternative approaches that enable users to iteratively construct and refine generated designs may be needed especially for animations.”
How does the tool work?
Imagine transforming a flat illustration of a rocket blasting off into a vibrant animation with fiery trails and starry skies. That’s the core strength of Keyframer. Users upload an image (specifically, Scalable Vector Graphics) and type a text prompt like “make the stars twinkle and the rocket soar into space.” The tool, powered by the powerful OpenAI GPT-4 language model, then generates CSS animation code that brings the image to life based on your description.
Keyframer offers multiple animation iterations from a single prompt, allowing you to choose the one that best captures your vision. You can further refine the animation by tweaking colour codes, animation duration, and even directly editing the generated code for more granular control.
The researchers also explained how the tool is different from image generating options like Dall-E. “While one-shot prompting interfaces are common in commercial text-to-image systems like Dall·E and Midjourney, we argue that animations require a more complex set of user considerations, such as timing and coordination, that are difficult to fully specify in a single prompt—thus, alternative approaches that enable users to iteratively construct and refine generated designs may be needed especially for animations.”
How does the tool work?
Imagine transforming a flat illustration of a rocket blasting off into a vibrant animation with fiery trails and starry skies. That’s the core strength of Keyframer. Users upload an image (specifically, Scalable Vector Graphics) and type a text prompt like “make the stars twinkle and the rocket soar into space.” The tool, powered by the powerful OpenAI GPT-4 language model, then generates CSS animation code that brings the image to life based on your description.
Keyframer offers multiple animation iterations from a single prompt, allowing you to choose the one that best captures your vision. You can further refine the animation by tweaking colour codes, animation duration, and even directly editing the generated code for more granular control.