Leading hair for the Meta Avatars 2.0 launch.

Getting familiar with the style

Before the launch, I did visual development for the new Meta Avatars 2.0 style, building character maquettes to learn its shape language firsthand. Vis dev is where you chase appeal and possibility, free of constraints.

The launch demanded the opposite discipline, which is what everything below is about.

The mission

I took the wheel on a launch with a new team, a concurrently-evolving technical foundation, and a tight timeline. Two problems at once: cover everyone’s hair in the new style at launch, and stand up the production system (pipeline, visual direction, vendors, team workflows) to actually deliver it. The migration had to feel like you, not a reset.

Before deciding what to make, I mapped where these avatars actually live: different renderers, fixed budgets, reading clearly at both high and low resolution. That whole-system view is what let me make the right calls later.

• Family of Apps
• VR headset
• Companion app · gameplay
• Companion app · editor
• Stickers · real-time, low budget
• High & low resolution

Challenge · plan for the long term

The representation framework

I built a representation framework: a categorization system so “representation” became measurable, every style a point across a few axes. It did double duty. It defined the set catalog of close to a hundred styles we had to deliver by the Oct 2024 launch, and it gave us the map to plan continued representation long after.

• Hair type
• Length
• Style
• Hairline
• Unique
• Facial hair
• Hats
• Colour

Once every style had coordinates, the catalog stopped being a list. Patterns that made hair reusable became visible, and so did the thin spots.

The catalog, organized by the framework

One axis, in full

Hair type the axis representation turns on

Eight types, each a coordinate on one axis. Length, style, hairline, facial hair, hats and colour work the same way. Real catalog renders, asset codes removed.

Challenge · define what “done” means

Coverage that resonates, not parity

At launch, coverage looked backward. The target was the legacy catalog: every person who already had a look had to find themselves in the new style on day one. That version of coverage is measurable, because the thing you are matching already exists.

So I defined what done meant. Rather than counting one-to-one replacements, I made per-asset quality the bar: for each legacy look, an existing user had to find a new style they genuinely recognised as theirs.

Tracking coverage by category is what made the thin spots legible. Counting styles per category showed where the catalog was crowded and where it was sparse, so a gap was a shape in the framework rather than an opinion in a review. Some of those we closed inside the launch set itself, by reworking existing styles rather than adding new ones.

All of it was a prediction. We had no usage data yet, so the framework was our best read on where representation would fall short. Checking that read is what the year after launch was for.

Challenge · art production within hard constraints

Making it beautiful inside the budget

Every structural lever was locked: vertex budgets, LOD ranges, texture, rig, and shader limits. So I pushed the levers that were left: texture, shading, and above all a simple-but-interesting silhouette, the real signature of the new style. When a hair-simulation feasibility question threatened the look, I built comparative test assets and documented the trade-offs, turning an art-vs-Tech Art stand-off into a shared, evidence-based call that held aesthetic intent within budget.

And every hairstyle had to sit cleanly under every hat, across four headwear categories, a manual, repetitive step that was quietly throttling the whole team. I spotted the bottleneck, framed the problem, and worked with our Tech Artist on a fix: they built a procedural mesh-generation tool, I was its first user, and we iterated until it reliably automated the step. Then we rolled it out to the rest of the artists. Constraints weren’t the ceiling. They were the brief.

Constraints, solved in the asset

Under hats every style, deforming cleanly

Colour one mesh, the full system

Challenge · scale the team & throughput

Building the team that built the catalog

I turned a stalled start into a repeatable machine, and a team. I led and grew the people who made the catalog: concept artists, 3D artists, tech artists, a producer, and external vendors. A 3D style guide captured every style in the framework; artists mentored into independent sub-stream ownership; and vendors scaled to the majority of output through codified quality rubrics, batch reviews, and documentation, giving throughput without added review overhead. I kept briefing the why, because motivated people do their best work.

• Concept artists
• 3D artists
• Tech artists
• Producer
• Vendors

Outcome · what shipped

Launch

The new-style avatar system in players’ hands.

0 → 90%

of the legacy catalog covered, in one half, so existing users found their look on day one

5 apps

Facebook, Instagram, WhatsApp, Messenger and Horizon

Still live

the launch catalog is what the product ships today

Decision · stop scaling by hand

The 2025 push, and where it hit a wall

Once people were in the app, the pre-launch predictions could be checked against real use. Some gaps we had called accurately. Others we had not anticipated, and people asked for them. Both landed on the same map, and the goal changed shape: no longer covering what the legacy catalog held, but growing toward the likeness of billions of people, with no existing set to measure against.

What became clear is that gaps are not all the same kind of problem. Sorting them turned one broad ambition into three specific asks, each with a different owner.

The third kind did the most work. A technical gap, expressed as the representation it is blocking, is a far stronger case for investment than the same gap expressed as a missing capability.

In 2025 I led the representation expansion, deepening the textured, coily, and culturally-specific styles that both research and real usage showed were still missing. It surfaced a hard limit: on the existing pipeline, every style was authored one hair at a time, each new look a fresh bespoke build.

That math doesn’t close a representation gap. The harder we pushed for coverage, the slower and costlier it got, and the long tail of still-under-served communities was exactly the part that cost the most to reach. Representation at scale needed a different foundation, not more hands.

Decision · build the system

The proposal: scale into a system

I made the case to stop building bespoke and start building a system: a modular, parametric architecture and tooling where a small set of well-designed, reusable parts compose into a large space of styles. Coverage stops scaling with headcount and starts scaling with the system: more representation, faster, at lower marginal cost, with quality held by design.

Getting it built took more than an art argument. I authored a hair strategy that synthesized UXR gap analysis, PM product goals, and producer budget into a single, fundable plan, translating a craft problem into business terms leadership could back. It secured funding and a roadmap, turning “make more hair” into “build the thing that makes the hair.”

Decision · design for creators

Putting it in creators’ hands

The same modular parts that let us scale become the building blocks creators compose with. The foundation was designed to go further: tooling that lets creators customize and shape their own looks. My focus was the creator experience: how that tooling would be used, and what customization should feel like. I partnered with PM to understand the economy framework it would live inside, and what kinds of assets creators might make, so the tooling was designed with a sustainable creator economy in mind.

Shared here as direction and thinking, built on the shipped work above: the how, not the what.