The room is empty. Then it isn't.

A camera company wearing an interior-design costume

Walk into any open house this year and there is a decent chance the photos that lured you there were never staged with real furniture. The sofa, the rug, the moody twilight sky behind the windows - all of it generated by Collov AI from a single picture of an empty box of a room. The agent uploaded a photo, picked a style, and a credible home appeared.

Collov AI - the public face of Collov Labs - sits in Redwood City and calls itself an interior-design platform. That is true and also slightly beside the point. The interior design is the part you can see. Underneath is a company building systems that look at images and figure out what to do with them, which is a much stranger and more ambitious project than picking out throw pillows.

The reported numbers are not small for a company this size: more than 6 million users across 200-plus countries, over 300 business customers, and roughly 39 employees holding it all together. In April 2026 they closed a $23 million Series A and opened a research lab. For a business that started by digitally arranging furniture, that is an unusual amount of momentum.

Staging a room cost too much and took too long

Here is the tension that the whole company hangs on. A furnished listing sells faster and for more money - this is not controversial, agents have known it for decades. The trouble is that physical staging is expensive, slow, and logistically miserable. You rent furniture, you hire movers, you schedule a photographer, and you do it for one house while ten others sit empty and unsold.

Empty rooms, it turns out, are bad at selling themselves. Buyers struggle to imagine a life in a blank rectangle. They under-bid, they scroll past, they lose interest. The gap between a vacant room and a desirable one was made of money and hours, and almost nobody had a fast way to close it.

You could, of course, hire a person in Photoshop to fake the staging by hand. People did. It cost a few hundred dollars and took a few days per home, which solved the furniture-truck problem by inventing a slower, pricier one. The industry had a workaround for its workaround. That is usually the sign that someone is about to build something better.

A physics PhD decided rooms were a vision problem

Xiao Zhang did not arrive from the furniture world. He came out of Stanford with a PhD in applied physics, having written a thesis on using machine learning to optimize free-electron lasers - which is exactly as far from scatter cushions as it sounds. He was named to the Forbes 30 Under 30 list. Then he started a company about decorating rooms, which on paper looks like a detour and in practice was the point.

The bet Zhang and co-founder and CTO Casey Zhou made was that interior design is not really a taste problem to be solved by a catalog. It is a vision problem - a question of teaching software to understand the geometry of a space, the light falling across it, and where a chair can plausibly go. Solve that, and staging a listing is just one thing the software can do. The team they built leans academic, pulling researchers from Stanford, UC Berkeley and Yale.

The approach combines diffusion models, spatial reasoning, and multi-step agentic systems - software that can look at a scene and then carry out a sequence of edits, not just generate one pretty picture. Most companies would have been content to be a good staging tool. Collov treated staging as the demo and visual intelligence as the actual product.

Point a phone at a room. Get options.

The mechanics are deliberately boring to use, which is the highest compliment you can pay software like this. Upload one photo of an empty or lived-in room. The system returns a fully furnished, photorealistic version in roughly ten seconds. Don't like the style? Pick another. Want the existing sofa gone? Erase it. Different floor, different walls, twilight sky instead of grey afternoon? All a click.

The CozyAI app is where the camera thesis gets literal. Strap on an Apple Vision Pro, stand in your actual living room, and redecorate it in front of your own eyes - no screwdriver, no regret, no second trip to the store. It is a glimpse of where the company thinks all of this is heading: not a website you visit, but a layer over the world you are already looking at.

The pixels are doing the selling

A vision is only as good as the listings it moves. Collov's own reported figures point at the part of the funnel agents actually care about: more qualified buyers, faster sales, higher prices. Treat these as the company's numbers rather than an independent audit - but they are the case it makes to the people writing the checks.

The other proof point is who is shipping Collov's models. The company says its visual AI is integrated into Qualcomm on-device systems - meaning the redesign can happen on the chip in your hand rather than in a distant data center. That is the kind of partnership you get when your software is genuinely about seeing, not just about sofas.

Teach AI to see, then to act

The $23 million Series A in April 2026 was not really a furniture round. It funded a research lab, and the lab's stated job is to build visual intelligence systems for the next generation of devices - wearables, real-time interactive hardware, the things you look through rather than type into. The investors - Matrix Partners, MindWorks Capital, Brightway Future Capital and Taihill Venture - were buying the camera thesis, not the catalog.

Strip away the real-estate framing and the mission is simple to state and hard to build: make AI that understands images well enough to do something useful with them, automatically, in sequence. A room is a convenient first test case because the stakes are visible and the payoff is obvious. But a system that can read a room can, in principle, read a shelf, a street, a workbench, a face.

Back to the empty room

For most of computing history, talking to a machine meant typing. Then it meant talking. Collov is wagering that the next step is showing - that you will point a camera at the world and the software will understand the world and then change it for you. If that bet lands, virtual staging will look like the quaint first chapter of something much larger, the way the calculator app looks next to the phone it lives on.

There is a healthy reason to stay skeptical. The uplift numbers are self-reported. The competition - Virtual Staging AI, REimagineHome, Interior AI and a crowd of others - is real and growing. "Visual intelligence" is a phrase that can mean everything and therefore nothing. Collov has to prove that a staging tool can become an AI platform without losing the customers who just wanted a nicer listing photo.

So return to that open house. The room was empty - a blank rectangle that nobody could picture living in. An agent uploaded one photo, and ten seconds later it had a sofa, a rug, warm light through the windows, and a price that crept upward. The furniture was never there. The buyer who fell for it never noticed. That is the whole company in one room: Collov AI didn't move a single chair, and somehow it staged the house anyway.