Why Virtual Reality Avatars Need More Than a Realistic Face
A virtual person can look convincing in a screenshot and become distracting as soon as it moves. A useful VR avatar needs coherent movement and a clear role in the experience. When it represents your own body, how it responds to your actions matters as much as its appearance.
The 2020 paper The Rocketbox Library and the Utility of Freely Available Rigged Avatars examines this work behind the scenes. It explains how digital bodies are built and animated, then reviews their use in research on virtual experiences. Read the paper.
What a rigged avatar actually is
The visible surface of an avatar is only one part of the model. A rig adds an internal control structure, often described as a skeleton, that lets software pose and animate the body.
That is what allows a digital person to turn, reach, walk, or make an expression without rebuilding the model for every frame.
The distinction is easy to see in an imagined example. A beautifully modeled hand that bends incorrectly can be distracting. A simpler hand that follows a movement naturally may serve the experience better. Appearance and behavior need to work together.
The Rocketbox paper treats avatar creation as a sequence of design decisions, from the shape of the body to the way movement is captured and applied.
Why a virtual body matters to research
An avatar may represent another person in a shared environment, a computer-controlled character, or the user's own body.
That last use is central to embodiment research: investigating the experience of a virtual body as belonging to oneself. It gives researchers a way to vary features of a body or its movements while studying perception and behavior.
A reusable library can make those experiments easier to build and compare. Teams can begin with existing characters instead of commissioning every model from scratch.
However, providing the models does not validate every application made with them. A library used in rehabilitation research is a research resource, not evidence that any particular VR treatment improves health.
The paper and the library have different timelines
The original article describes the library's 2020 release and its capabilities at that time. Those details should not be mistaken for current product documentation.
As checked in November 2026, the official Microsoft repository describes 115 avatars and an MIT license. It also states that the repository was archived on October 2, 2026. Anyone evaluating it for a new project should read the current files and account for its archived status. See the official repository.
Look beyond the screenshot
When assessing an avatar experience, watch it move. Does the body respond as expected? Do its proportions work for the task? Are facial expressions appropriate? Can the intended users find representations that suit the experience?
These questions are more informative than realism alone. A virtual person has to behave convincingly over time, not just look convincing in a promotional image.
Ask for a moving demonstration
If you are choosing avatars for a project, test the movements your users will actually perform before judging the model on appearance. Include the people who will use the experience in that review. For a research project, document the avatar and animation choices so others can understand what participants saw.