Conversational AI Agentic Orchestration Vision AI AI Character Interior Design

Meta Reality Labs

Atelier
Luma

2025 · Proof of Concept · Mixed Reality Scroll to explore ↓
Overview

There's a moment in every great design collaboration where a room stops being what it is and starts becoming what it could be. Atelier Luma was built to create that moment — for anyone, in any room. A conversational AI experience that sees your space, understands your taste through conversation, and proposes changes in place. Built for Meta Quest Mixed Reality. Designed by Astral City.

The Challenge

Visualisation without direction is just a prettier problem.

Most people never get to work with a great interior designer — not because the desire isn't there, but because the process has always demanded the right contacts, the right budget, and a vision already half-formed. Digital design tools have closed some of the gap, but they start from a catalogue, not a conversation. Previewing isn't designing. Meta Reality Labs asked Astral City to prototype the harder thing.

The Solution

The relationship, the vision, and the room — all in the same moment.

Four systems, each feeding the next: conversation to surface taste and direction, spatial awareness to read the room, design intelligence rooted in real interior design frameworks to propose changes in place, and iteration to refine them until the room feels done. The pipeline begins with a 16-point gaze-guided room scan — 16 photos processed by a vision AI to produce a structured spatial profile that underpins every observation, suggestion, and proposal Luma makes.

What We Built
01 — Arrival

A presence, not an interface.

Atelier Luma appears as a warm iridescent orb, orienting to the space and introducing itself. The character's voice, form, and behaviour were crafted to feel considered — because in Mixed Reality, credibility is earned through presence, not prompts.

02 — Room Scan

Seeing the room is the prerequisite for understanding it.

A 16-point gaze-guided capture feeds a vision AI that builds a structured spatial profile of the room. Every observation, suggestion, and proposal Luma makes is anchored in what it has actually seen — spatial grounding before any design suggestion is made.

03 — Dialogue

The brief emerges from the exchange, not from a catalogue.

An LLM conversation engine surfaces taste and direction — asking about colour, mood, material, and lifestyle to build a design brief through natural conversation. Dialogue is the design interface. An intent classification layer keeps conversation natural while managing state transitions underneath.

04 — Iteration

The highest-value interaction isn't the reveal. It's the loop.

A selected item triggers the in-painting pipeline — grounded in real interior design frameworks, visualised as a floating image card in the room, and refined through live feedback until it's right. V2's pivot to real-time furniture iteration defined the entire V3 roadmap.

From the Studio

"Luma is a designed personality as much as a technical system — credibility in Mixed Reality is earned through presence, not prompts."

Tech Stack
Development Engine Unity
Hardware / OS Meta Quest MR / Horizon OS
Intelligence Vision AI / LLM Conversation
Visual Generation In-painting
Infrastructure Backend API / AI Back Office / Cognitive3D
Outcomes

Two prototypes. One sharp pivot. A roadmap built on what we found.

V1 proved the architecture — room scan, moodboard, design reveal. V2 pivoted around a single finding: the highest-value interaction was a person and Luma, iterating on furniture proposals in real time. Everything sharpened around that loop. The process surfaced five concrete technical learnings — latency, spatial capture, AI presence, positioning, and visual output — each documented with a specific recommended future state.

V3 Roadmap
  • Latency: target response to under 2 seconds end-to-end
  • Spatial capture: higher fidelity room understanding for more precise placement
  • AI presence: richer physical behaviours and ambient motion states
  • Positioning: smarter spatial awareness for more natural conversation proximity
  • Visual output: higher quality in-painting with better material accuracy
Why This Matters

Atelier Luma demonstrates what becomes possible when a designer's process, spatial awareness, and AI meet in the same room — a relationship built through conversation, realised in place, and followed through from first proposal to delivered room. The most telling moment isn't technical. It's when Luma drifts across the room, pauses beside the sofa, and says it can already see the beginnings of a direction — and the room, for the first time, feels considered.

Ready to build something like this? →
Capabilities
Conversational AI Design Vision AI Integration AI Image Generation AI Character Design Mixed Reality Development Rapid Prototyping

Next Case Study

World Play Day →