AuRA Avatar Generator

Automated video briefings need someone on screen, and the right presenter depends on the audience. The AuRA Avatar Generator is our companion tool for AuRA (Autonomous Research Assistant): it creates customizable avatars and avatar-driven video clips that the pipeline can then lip-sync to a narrated script.

What it does

• Builds a presenter from attributes. Dropdowns set age, gender, ethnicity, hairstyle, clothing, accessories, lighting, background objects and scene context, so a presenter can be matched to a specific audience, from patient education to a professional conference.
• Generates video, not only stills. Resolution, frame rate and clip length are adjustable, and the tool can output a single image or a full sequence.
• Runs inside the pipeline or on its own. A Gradio interface handles interactive design, and a command-line entry point lets the research pipeline request a presenter with a text prompt.

How it works

Text-to-video generation is based on Lightricks' LTX-Video: a video autoencoder, a 3D transformer and a rectified-flow scheduler, running on PyTorch. MoviePy handles assembly and Real-ESRGAN upscales the output. The interface turns the chosen attributes into a structured prompt, so producing a new on-screen host means changing a few settings rather than writing a prompt by hand.

The holiday greeting above, cut together from a range of generated presenters, shows the variety a single tool can produce. For where the presenters go next, voices and lip sync, see AuRA's Voices and Avatars page.