Why Does an Avatar Move?
Have you ever wondered how a virtual character's expressions, movements, and lip-sync actually work?
Most people say "AI drives it." But AI is just the decision-making layer. Before AI can determine "this person is happy," there needs to be a system that translates "happy" into a language a 3D model understands. That language is VRM Animation, or VRMA for short.
Let's break it down.
01 · What Is VRM?
VRM (Virtual Reality Model) is a 3D model format purpose-built for virtual characters, built on the glTF 2.0 standard, developed by the VRM Association under Japan's pixiv. It has become the de facto standard for VTubers, VRChat, and virtual livestreaming.
What made it a standard is that it packages everything a virtual character needs:
- Skeleton system: 57 standard humanoid bone nodes with automatic model adaptation
- Material system: Supports MToon (toon shading), PBR (physics-based rendering), Unlit (no lighting), and more
- Expression system: BlendShape supporting both preset and custom expressions
- Physics simulation: SpringBone for natural hair and clothing movement during motion
02 · What Is VRMA?
VRMA (VRM Animation) is a format specifically designed to describe humanoid animations for VRM models.
Its core value in one sentence: The same animation file works on any VRM model.
Without VRMA, if you record a dance animation for one specific VRM model, you'd need to re-record or manually retarget it for every different VRM model. But with VRMA — it's universally compatible across models. You bind it to a target model's skeleton system and it just works.
This is the key difference between VRMA and ordinary video or proprietary animation formats: It's an open, universal language for humanoid animation.
03 · What's Inside a VRMA File?
A VRMA file contains three types of information:
Skeleton animation: Describes rotational movement of bones — raising arms, turning, walking — using glTF's humanoid skeleton structure to automatically adapt to the target model's bone topology.
Expression animation: In VRM, expressions use BlendShape (morph targets). BlendShape stores the vertex position delta between a "neutral" and "target" shape. By adjusting the weight value (between 0 and 1), expressions smoothly transition from "neutral" to "laughing."
Eye-tracking animation: VRM models have dedicated eye-tracking bones. VRMA can precisely describe how gaze direction transitions between different orientations.
04 · How Is Animation Made?
The most common pipeline:
Step 1: Get raw animation from AI or motion capture. AI tools (like Mootion) can generate FBX animation files from text descriptions. Or use a motion capture device to record directly in BVH format.
Step 2: Convert to VRMA format. FBX or BVH needs conversion. Blender handles FBX-to-BVH conversion, then Unity with the UniVRM plugin completes the final conversion. The most critical part: T-Pose detection — the BVH file's initial pose must be a standard T-Pose, or bone mapping will fail.
Step 3: Import into your application. The resulting .vrma file can be dragged directly into any supporting application.
05 · Which Tools Support VRMA?
| Type | Tool | Notes |
|---|---|---|
| Unity Plugin | UniVRM | Official standard implementation |
| Web Rendering | three-vrm (@pixiv) | Browser-side VRM rendering |
| Blender Plugin | VRM Add-on for Blender | Native Blender support |
| Online Conversion | bvh2vrma | Pure web tool |
| Platform | VRoid Hub | Model and animation distribution |
| Virtual Desktop | VMagicMirror | Windows virtual character app |
| Steam App | VRM Posing Desktop | Pose creation tool |
Additionally, Mates.ink — the Chrome browser virtual character platform — fully supports VRMA format, allowing VRM animations to load and play directly in the browser.
06 · What Can FEAIA Do Now?
FEAIA currently drives Avatar reactions through MatesScript.
MatesScript is FEAIA's animation description language, using a preset emotion state machine (EmotionStateMachine) to drive Avatar expressions and camera movements. When AI determines "this question is interesting," the Avatar automatically triggers a "joy" emotion, showing the corresponding expression and camera movement.
This is a lightweight, signal-based animation system — fast, immediate, requiring no full animation files.
In the future, FEAIA will offer full VRMA animation support for premium members. This means:
- Import professionally motion-captured dance, tutorial, and daily life animations
- The same VRMA animation works on any VRM Avatar
- Ecosystem integration with platforms like Mates.ink, sharing animation assets
07 · Why Is This Worth Anticipating?
The virtual character industry has a long-standing pain point: Animation production is extremely expensive. A 30-second professional VTuber animation can cost thousands of dollars. One major reason: animations are tightly coupled to specific models — animations made for Model A can't be used on Model B.
VRMA solves this. It transforms animation into an independent, reusable asset — just like a font file works in any software, a VRMA animation plays on any VRM model.
When AI can generate these animations, the barrier drops even further.
Today, tools already exist that generate animations from text descriptions. Combined with VRMA's universality, an independent creator with no motion capture equipment can still generate custom animations for their Avatar. FEAIA is moving in this direction.
This article explains the fundamentals of VRM and VRMA technology to help readers understand how Avatar animation works. FEAIA's current features are powered by MatesScript. Full VRMA support is a future feature for premium members.
Mates.ink (Chrome extension) already supports VRMA format.
