Inside Our AR Interior Design App: Staging, Walls and Floors

Interior AI started with a simple promise: take a photo of any room, pick a style, and see it redesigned in seconds. Since launch, our users have kept asking the same follow-up question: "That looks great, but will it actually fit in my room?"

A beautiful AI image answers "what could this look like?" It doesn't answer "what will this look like here, in my space, at my scale, in my light?" That second question needs augmented reality. This is the story of how we added AR staging and live wall and floor design to Interior AI, and the engineering decisions behind it.

From a FlutterFlow MVP to a native AR app

The first version of Interior AI, launched in 2024, was built with FlutterFlow. It was the right call at the time. We validated the idea fast, shipped to the App Store, and learned what people actually used.

But AR is where cross-platform abstractions hurt the most. Plane detection, scene reconstruction, occlusion and lighting estimation all live deep inside Apple's frameworks, and they need to run every frame without dropping a single one. So in early 2026 we built two focused AR prototypes to prove the hardest parts: one for detecting and re-surfacing floors, and one for turning a room into a navigable 3D model. Once they worked, we rewrote Interior AI natively in Swift and SwiftUI and folded those lessons in.

The result is a single app where AI generation and AR aren't separate features. They feed each other.

AR staging: real furniture, real scale, your room

AR Staging lets you place 3D furniture into your room through the camera, at true size, and walk around it.

Under the hood it runs on ARKit and RealityKit. The app detects both horizontal and vertical surfaces, so you can place a sofa on the floor, a mirror on the wall, or a pendant lamp on the ceiling. We let you choose the surface type explicitly instead of guessing. In testing, automatic detection almost always returned "floor" even when you were pointing at a wall, and nothing breaks the illusion faster than a painting that falls to the ground.

A few details make the difference between "a 3D model floating on my screen" and "a chair in my living room":

• Real-world lighting. The app reads ARKit's light estimate every frame and adjusts the virtual light's intensity and direction to match your room, so a chair near a window doesn't look lit by a studio lamp. We also use environment texturing, so reflective surfaces pick up your room's actual colors.
• Occlusion on LiDAR iPhones. On Pro models with LiDAR, the app reconstructs a mesh of your room. Real objects then hide virtual ones: the coffee table you place behind your real sofa actually disappears behind it.
• Correct pivots. 3D models come from many sources, and their origin points are often in odd places. We normalize every model so floor items sit exactly on the floor, wall items sit flush against the wall, and ceiling items hang the right way up.
• Natural gestures. Tap to select, drag to move along the surface, pinch to scale, twist to rotate, double-tap to reset to true size.

Where the furniture comes from

You can choose from a curated library of more than 2,700 ready-to-place 3D models: sofas, tables, chairs, shelving, lighting and décor, grouped by where they go. Each model downloads once and is cached, so the second placement is instant.

You can also bring your own. Image to 3D turns a single product photo into a textured 3D model using TRELLIS, an open image-to-3D model we run on our own GPU servers, so you can stage the exact armchair you're eyeing in an online store. And Object Capture uses Apple's on-device photogrammetry to scan a real object around you into a 3D model, with no LiDAR required.

Live AR wall and floor design

The feature that surprised our users most is AR Surface Design: point your phone at a room and change the floor and walls live. Swap tile for oak, paint a wall terracotta, try marble. You see it on the real surfaces as you move.

Detecting surfaces without guessing

A common approach is to run an image segmentation model on every camera frame and recolor the pixels it labels "wall." It looks impressive in demos but flickers at the edges and breaks when the camera moves. We took a geometric approach instead. ARKit detects the actual planes in your room, and we cover each one with a real 3D surface that grows as ARKit learns more of the room.

The hard part is deciding which planes are really your floor and walls:

• Floors. When the device can classify surfaces, we accept only planes labeled as floor and reject tables, seats and ceilings. When classification isn't available, we track the lowest horizontal surface and treat anything within a few centimeters of it as floor.
• Walls. A kitchen counter, a fridge and a wall are all vertical surfaces. We only accept a vertical plane as a wall if it spans at least two meters in height and is wide enough to be a wall. That threshold came from real-world testing: a fridge is about 1.8 m, and it kept getting painted.

This approach works on non-LiDAR iPhones too. LiDAR adds better occlusion, but you don't need a Pro phone to redesign your floor.

Materials that look physically right

Colors are simple. Textures are where AR usually looks fake. Every texture in Interior AI is a full physically based material with color, surface normal, roughness and ambient occlusion maps. That's why oak shows grain under your lamp and polished marble catches reflections while concrete stays matte.

We also tile every texture at real-world scale. A 30 cm tile looks like a 30 cm tile whether you're standing over it or looking across the room. The library currently has 247 textures (115 floors and 132 walls) across wood, tile, marble, plaster, brick, concrete, wallpaper and more, plus a 12-color palette. You can also use a photo of your own sample.

Keeping it at 60 fps

Loading high-resolution textures can freeze an AR session for a fraction of a second, which is enough to feel broken. We load all texture maps in parallel off the main thread, apply them together in one step, and cancel any in-flight load the moment you pick something else, so the last tap always wins.

When AI meets AR

The most useful part isn't AR or AI on its own. It's how they connect:

• Snapshot to staging. Take a snapshot of your AR design and send it straight into AI Virtual Staging to generate a fully furnished, photoreal version.
• Spatial results. Every AI design can be viewed as a "spatial" image. An on-device depth model estimates the scene's depth, and the image gains a subtle 3D parallax as you tilt your phone. It runs entirely on your iPhone and respects Reduce Motion and Low Power Mode.
• Room scanning. On LiDAR iPhones, Room Scan uses Apple's RoomPlan to capture your room as a 3D model, with a 2D floor plan and PDF export you can hand to a contractor.
• Floor plan to 3D. Upload a floor plan image and Interior AI turns it into a 3D space you can walk through.

What's next

We're currently revamping Interior AI's UI and UX to make these tools even easier to discover, and we're building the Android version, which is coming soon. Interior AI is also available on the web at interior-ai.app, where you can try it for free.

If you're redesigning a room, download Interior AI on the App Store and see your ideas in your own space before you buy anything.

And if you're building a product that needs AR, on-device AI or both, talk to us. Everything in this article is the kind of work we do for clients too.