VRM (Virtual Reality Markup)

VRM (Virtual Reality Markup) is an open, royalty-free file format for 3D humanoid avatars, built as an extension of the glTF 2.0 standard. It defines a comprehensive schema for avatar data, including mesh geometry, materials, blend shapes for facial expressions, and spring bones for dynamic hair and clothing physics. The format's primary goal is to ensure avatar portability and interoperability across different VR/AR platforms, social applications, and game engines, preventing vendor lock-in. It is governed by the VRM Consortium, a Japanese organization promoting its development and adoption.

The specification addresses key challenges in avatar representation through several dedicated extensions. The VRM Humanoid extension defines a standardized bone structure, allowing avatars to be retargeted to different animation rigs. The VRM Blendshape extension controls facial expressions and lip sync via predefined morph targets like "Blink" or "Joy." Furthermore, the VRM Spring Bone system simulates secondary motion for accessories, while VRM FirstPerson settings manage how the avatar is rendered from the user's own viewpoint (e.g., making the head mesh invisible in first-person view).

VRM files are created using authoring tools like UniVRM for Unity, which provides exporters, importers, and utilities for validation. A typical workflow involves modeling and rigging a character in a 3D tool like Blender, then using the UniVRM plugin to add VRM-specific metadata, configure humanoid mapping, and define expression presets. This pipeline enables creators to produce avatars that are immediately functional in any VRM-compliant application, from VRChat to standalone VRM viewers and virtual meeting spaces.

The adoption of VRM is significant for the open metaverse ecosystem, as it provides a common avatar lingua franca. Unlike proprietary avatar systems, VRM's open specification allows for a decentralized creation economy where users can own, customize, and transfer their digital identity. Its reliance on glTF ensures wide support in modern graphics pipelines and web-based environments using frameworks like Three.js. This technical foundation makes VRM a critical standard for enabling user-generated content and social interaction in interoperable 3D worlds.

VRM (Virtual Reality Markup) is a file format and scene description language, originating in the mid-1990s, designed to create interactive 3D worlds viewable through web browsers. It was developed as part of the Virtual Reality Modeling Language (VRML) specification, with VRM often used interchangeably with VRML in its early iterations. The core concept was to provide a text-based, human-readable markup—similar to HTML for web pages—to define geometry, lighting, and basic interactivity for virtual spaces on the nascent World Wide Web.

The origin of VRM is deeply tied to the VRML 1.0 specification finalized in 1995, following proposals from pioneers like Mark Pesce and Tony Parisi. It emerged from the Silicon Graphics (SGI) Open Inventor file format, with its syntax adapted for network transmission. The "Markup" component signifies its role as a declarative language where authors define a scene graph—a hierarchical tree of nodes representing shapes, transforms, and materials—rather than writing imperative rendering code. This allowed 3D content to be created and shared more easily across different platforms.

While revolutionary for its time, VRM/VRML was ultimately superseded by more powerful and efficient technologies. Its legacy is evident in modern standards like X3D (the official successor to VRML) and glTF, the contemporary "JPEG of 3D." The etymology of VRM highlights a pivotal, if transitional, phase in making 3D graphics accessible on the open web, establishing foundational concepts for scene description that continue to influence immersive media and metaverse development today.

The VRM specification is built upon the glTF 2.0 standard, a widely adopted 3D transmission format, and extends it with specialized extensions for humanoid avatars. At its core, a VRM file contains a complete 3D model with a defined skeletal structure, materials, textures, and the critical VRM metadata. This metadata is a JSON-based schema that defines avatar-specific properties, including human bone mappings, blend shape presets for facial expressions, first-person view configurations, and licensing information. This layered structure ensures compatibility with standard 3D pipelines while adding the semantic data needed for avatar interoperability.

The workflow for using a VRM avatar begins with import and validation. A compatible application, such as a game engine plugin or VR chat platform, loads the .vrm file, parses its glTF data, and validates it against the VRM schema. The system then maps the model's skeleton to a standardized humanoid bone hierarchy, allowing animations designed for one VRM avatar to work on another. Key features like blend shapes (for facial expressions like blink or smile) and spring bone physics simulations (for dynamic hair and clothing) are initialized based on the metadata, bringing the static model to life.

At runtime, the VRM avatar is driven by input data. This can include user tracking from VR controllers and headsets to control body and head movement, audio input to drive lip-sync visemes via blend shapes, or predefined animation states. The spring bone system calculates secondary motion in real-time, adding realistic jiggle and sway to specified parts of the model. For rendering, the avatar utilizes the MToon shader, a cel-shading style material commonly bundled with VRM, which provides a consistent, anime-inspired aesthetic across different platforms and lighting conditions.

A critical aspect of VRM's operation is its focus on creator and user permissions, enforced through its metadata. The format includes fields for authorship, contact information, and allowed usage (e.g., personal use, commercial use, prohibited behaviors). Applications can read this data to enforce license compliance automatically. Furthermore, the specification defines first-person view settings, allowing creators to designate which parts of the avatar's model should be rendered or hidden when viewed from the user's own perspective, preventing visual obstruction in VR.

The VRM extension for glTF is a formal specification that adds avatar-specific metadata and constraints to the core glTF (GL Transmission Format) 3D model standard, enabling the creation of portable, humanoid 3D characters for use in virtual reality, games, and social platforms. Defined by the VRM Consortium, this extension standardizes properties like blend shapes for facial expressions, spring bone physics for hair and clothing simulation, and first-person view camera configurations, ensuring avatars behave consistently across different compatible applications and engines.

At its core, the extension introduces a VRM top-level object within the glTF JSON, which contains all avatar-specific data. This includes the humanoid skeleton definition, mapping standard bone names (like hips, leftUpperArm) to glTF node indices, and material properties for advanced shading such as MToon, a cel-shaded style popular in anime-style avatars. The meta object within this structure stores crucial information like the avatar's name, author, licensing terms, and a reference thumbnail, making it self-describing and easy to catalog.

A key technical feature is the blend shape group system, which defines preset facial expressions (e.g., Joy, Angry) and viseme shapes for lip-syncing, going beyond the basic morph targets in standard glTF. The secondary animation system, often called spring bone, uses colliders and physics parameters to simulate jiggle dynamics for soft body parts, adding life to hair, tails, and accessories without requiring complex rigging or real-time simulation code in the host application.

For practical implementation, the VRM extension is supported by major tools and SDKs. Authoring software like VRM editor and UniVRM for Unity allow creators to export models from 3D applications like Blender into the .vrm file format, which is essentially a glTF 2.0 file with the VRM extension and embedded binary data. Runtime loaders, such as those provided by the VRM consortium, parse this data to reconstruct the avatar with all its humanoid, expression, and physics capabilities intact.

The standardization provided by the glTF VRM extension solves critical interoperability issues in the avatar ecosystem. It allows an avatar created for one social VR platform to be used in another, facilitates the development of avatar marketplaces, and provides a clear technical foundation for user identity in the metaverse. By building upon the widely adopted glTF standard, it leverages existing tooling and performance optimizations for 3D asset delivery while adding the specialized features required for expressive, interactive humanoid characters.