AI Avatar Stylist
A conversational AI that restyles your 3D avatar from a single sentence. Shipped 0 → 1 across Family of Apps and Horizon mobile.
Context
1B+ avatars created across Family of Apps and Horizon. The editor is the hub — where identity gets built, refined, and re-styled for every social surface users show up on.
Challenge
Two problems, one opportunity.
- The editor was overwhelming. Hundreds of items, nested menus, no guidance. Most users abandoned before finding a look they loved.
- Conversational AI had no prior art inside a 3D editor. AI was reshaping every creative tool, but nobody had figured out what it should feel like here. No patterns to borrow.
My role
End-to-end. From the first whiteboard to public launch at scale. I set the vision, defined the staged rollout, and ran the multidisciplinary team that built every phase.
- Defined the interaction model, IA, and design principles
- Scoped the staged rollout: Prototype → Dogfood → Public Test → Launch
- Directed supporting designers, art directors, 3D artists, and animators
- Owned AI naming + framing with brand and CDs
- Ran research partnerships with UXR — concept testing, usability, longitudinal
- Presented strategy to leadership and 100-person all-hands audiences
- Held weekly XFN office hours and creative reviews
Key design challenges
Setting expectations with AI
Two failure modes from early testing: users expecting magic, and users refusing to engage. Both kill the product. Fix: an honest NUX plus 100+ curated starter prompts. Capability taught by example, not by instruction manual.
Overcoming the blank canvas
An empty text field is the hardest surface in UX. Built a contextual suggestion system — starter prompts with personality, follow-ups that build on the user's last move. Creative momentum without telling people what to make.
Options, not answers
One AI output reads as take-it-or-leave-it. Built an "ingredients grid" instead — multiple outputs per prompt, mix and match between rounds. Users became co-creators rather than judges of one machine guess.
Bridging AI and manual control
Users wanted AI magic and fine-grained control. Designed a seamless mode-switch: AI restyles the silhouette, the manual editor tunes the details. Start with either, end with both.
Handling edge cases gracefully
Trust isn't built by what works — it's built by how the product handles what doesn't. About half of study participants arrived expecting full generative magic, and the system couldn't deliver all of it. Satisfaction held up after use anyway, because failure was treated as a first-class state instead of an edge case.
Proactive feedback — toasts and inline messages flag the moment a request isn't supported, so users aren't left guessing or repeating the same failed prompt. Seamless fallbacks — when AI can't deliver, the route to manual editing or an alternative flow is one tap away. No dead ends.
Public reveal
Style with AI debuted on stage at the Meta Connect 2024 developer keynote — the public commitment that locked in the launch.
Impact
400K+ new registrations to the Meta mobile gaming platform. 3rd-largest growth contributor across the entire product.
3× more spent on paid content vs. the traditional editor.
Ease-of-use scores beat platform benchmarks at launch.
Retention beat platform benchmarks — users came back.
Beyond launch
- Published 5 reusable AI UX patterns — adopted across other consumer AI teams at Meta
- Set the North Star for Prompt-to-Avatar (text · voice · photo → full avatar generation)
- Led 2026 AI Editing strategy planning with the XFN team
- Onboarded and mentored designers; established design ops templates used team-wide
As featured in
0→1 Product Design
AI/LLM UX
Vision Setting
Cross-Functional Leadership
UX Research Partnership
Interaction Design
Design Ops
Prototyping
Mentorship