Massive language fashions (LLMs) have the potential to influence a variety of artistic domains, as exemplified in well-liked text-to-image mills like DALL·E and Midjourney. Nonetheless, the applying of LLMs to motion-based visible design has not but been explored and presents novels challenges reminiscent of how customers may successfully describe movement in pure language. Additional, many present generative design instruments lack assist for iterative refinement of designs past immediate engineering. On this paper, we current Keyframer, a design device that leverages the code technology capabilities of LLMs to assist design exploration for animations. Knowledgeable by interviews with skilled movement designers, animators, and engineers, we designed Keyframer to assist each ideation and refinement levels of animation design processes by enabling customers to discover design variants all through their course of. We evaluated Keyframer with 13 customers with a spread of animation and programming expertise, analyzing their prompting methods and the way they thought-about incorporating design variants into their course of. We share a collection of design rules for making use of LLM to movement design prototyping instruments and their potential implication for visible design instruments extra broadly.