Metaverse avatar development: building avatars people actually want to wear

A metaverse avatar is a rigged 3D character that represents a person in a shared space, driven in real time by their headset, camera, keyboard or controller. Good avatar development balances four things: how expressive the avatar is, how much it costs to render, how much users can customize it, and whether it can travel between apps in an open format.

Avatars are where users form their attachment to a virtual space. People notice instantly when lip movement lags their voice, when hands clip through the body, or when the character looks nothing like the way they want to be seen. They are also a performance trap: one detailed avatar is easy, fifty on a standalone headset is not. This guide covers the decisions and pipeline behind avatar systems, from style choice to NFT-linked identities.

Choosing an avatar style

Style is a product decision with technical consequences.

Apple's Personas on Vision Pro and Meta's research on codec avatars show where photoreal telepresence is heading, but those rely on platform-level capture and are not something a typical app can replicate. Most products still do better with a well-designed stylized look.

Formats and interoperability

If avatars should work across multiple apps or survive a change of engine, pick open formats.

• glTF 2.0 / GLB from the Khronos Group is the baseline real-time 3D format, supported by every major engine and web library.
• VRM builds on glTF and adds what humanoid avatars need: a standardized bone mapping, expression presets, look-at settings, spring-bone physics for hair and clothing, and license metadata stating how the avatar may be used. It is widely used in the VTuber and social VR communities. The VRM specification is maintained by the VRM Consortium.
• OpenUSD is increasingly used in production pipelines and on Apple platforms, but it is less common as an end-user avatar exchange format.
• FBX remains common in authoring tools but is a proprietary format and a poor choice for runtime delivery.

Hosted avatar services that supply cross-app avatars through an SDK can save months, but several have changed terms, been acquired or shut down. If you depend on one, make sure you can export avatars in glTF or VRM and store them yourself.

The avatar pipeline

1. Base mesh design. Define proportions, style guide and a polygon budget. Modular bodies (head, torso, legs, hands) make customization and LOD easier.
2. Topology for deformation. Edge loops around joints, mouth and eyes so the mesh bends cleanly when animated.
3. Rigging. A humanoid skeleton compatible with your engine's retargeting system (Unity Humanoid, Unreal's skeleton, or the VRM bone map), with weight painting tuned at the shoulders, hips and fingers.
4. Facial setup. Blendshapes (morph targets) for visemes, the mouth shapes used for lip sync, and for expressions. Apple's ARKit set of 52 facial blendshapes has become a de facto reference for face-tracking input, so supporting it makes integration with tracking hardware easier.
5. Materials and textures. Physically based materials for realistic styles or toon shaders for stylized ones, texture atlases to reduce draw calls, and compressed formats such as KTX2.
6. Levels of detail. Two to four LODs, with the lowest suitable for crowds at distance. Impostors (pre-rendered sprites) can stand in for very distant avatars.
7. Customization system. Swappable meshes, color parameters and attachment points, defined as data so new items can be added without app updates.
8. Validation. Automated checks for bone names, triangle count, material count, texture size and scale before any avatar or item reaches users.

Driving the avatar in real time

Body tracking and inverse kinematics

A VR headset typically tracks the head and two hands or controllers. Everything else is inferred. Inverse kinematics (IK) solves for elbows, shoulders and spine from those three points, and procedural animation or machine-learning models estimate the legs. Meta's SDK offers body-tracking and generative leg estimation on Quest; on other platforms you will use your engine's IK or a third-party solver. Expect to spend time tuning edge cases like crouching, reaching behind the body and sitting.

Face, eyes and lip sync

Where hardware supports it, face and eye tracking can drive blendshapes directly, which greatly improves presence. Without it, audio-driven lip sync maps speech to visemes, and procedural blinking and idle eye movement prevent the "dead stare" effect. Always let users turn face and eye tracking off, and treat that data as sensitive.

Desktop and mobile users

Flat-screen users need a different control scheme: walk, emote and gesture with keyboard or touch. Avatars should look natural in both modes, and many platforms use webcam-based face tracking as an optional upgrade.

Performance budgets

Avatars usually consume the largest share of a social scene's rendering cost, because they are skinned, animated and unique. Starting points that many teams use for standalone headsets:

• A per-avatar triangle budget in the low tens of thousands at the highest LOD, much less at distance.
• One or two materials per avatar, with textures packed into atlases.
• A cap on bones per avatar and on spring-bone chains for hair and clothes.
• Animation and IK update rates reduced for distant avatars.

Treat these as budgets to confirm by profiling on your weakest target device with your expected peak number of visible avatars. If users can upload their own avatars, enforce budgets automatically and fall back to a simplified version when an avatar exceeds them.

Customization and the creator economy

Customization drives engagement and, for many social platforms, revenue. The design questions are practical:

• Who creates items? Your art team only, approved creators, or anyone? User-generated content needs validation pipelines, moderation for offensive designs and a takedown process for copyright claims.
• How do items fit different bodies? Shared body templates or automatic fitting systems avoid clipping when a jacket meets an unusual body shape.
• How are items sold? In-app purchases, a marketplace, or limited drops.

NFT avatars: when they make sense

In 2021–2022 many profile-picture NFT collections promised 3D avatars usable "across the metaverse." Few delivered usable cross-app avatars, partly because every app has different style, rig and performance requirements, and partly because some collections were abandoned. If you do want avatars or wearables as NFTs:

• Use the NFT as an ownership and license record, with the token metadata pointing to a glTF or VRM file stored on durable storage such as IPFS with pinning or Arweave.
• Be explicit about the license: can holders use the avatar commercially, in any app, or only yours?
• Accept that interoperability requires partner apps to support your format and style; a token does not make that happen by itself.
• Use ERC-1155 for wearable editions and ERC-721 for unique characters. Background on the collection side is in the guides to BAYC-style collections and CryptoPunks-style projects.

For most products, an off-chain avatar system with optional NFT wearables is the safer design.

Privacy, safety and inclusion

• Face, eye and body tracking data can reveal health conditions and emotional states. Minimize what you collect, process on-device where possible, and disclose clearly.
• Scanned or AI-generated likenesses require consent. Prevent users from creating avatars that impersonate real people.
• Offer diverse body types, skin tones, hair textures, assistive devices and cultural clothing. Users notice when they cannot represent themselves.
• Give users control over personal space and who can see their avatar up close.

Cost drivers

As a rough illustration, a stylized avatar system with a few body types, a facial rig and a launch catalog of a few dozen items might take one technical artist, two 3D artists and one engineer three to five months. A photoreal pipeline or a full UGC marketplace is a much larger project. Avatars sit inside a bigger application; see metaverse application development for the surrounding architecture and social metaverse platforms for where avatars matter most.

Frequently asked questions