Inspired by the developers’ creativity with LLM Agents, and the latest developement on multi-modal models, the file format and SDK is designed with these 2 design principles in mind.

  1. Lightweight and self-contained. Character visuals are designed to be run efficiently in massively parallel environments, and are deeply integrated with .
  • Switching styles, voice and LLM models should be as easy as switching apis.
  • Everything is standardised with the simplest abstraction in human and agent-readable format.
  • Avatars’ visual animation would be a set of clean instructions usable by Agents’ capability over time. The animation itself is self-contained without extra dependency.
  • Avatar would be able to react to non-textual input as well, for example eye gaze, scrolls and other audiovisual inputs
  1. Clean and simple interface. All designs should priortise on simplicity and have semantic based functions, from which users can compose complex character behaviours
  • toolings and interface should be supported across vector & rasterised (pixels) assets concurrently.
  • we provdie off-the-box animation pack when agent is using different tools and components, while keeping the framework flexible enough to stack new animations
  • We balance the need of character designer, artists and develoeprs such that they have a common ground to work together. We abstract the complexity of using avatars across styles and form factor on flat screen.
Intro to Avatars SDKAvatar Creation