X Debate: levelsio Ships Video-to-3D Rooms in Interior AI in 3 Hours, Turning a Walkthrough Video Into a Walkable, Editable 3D Space With Self-Hosted Headless Blender

Background and Industry Pain Points

For the past few years most AI interior design apps have followed the same pattern: upload a room photo and an image model returns a redesigned picture. It looks good, but it is only a picture. You cannot walk through it or move furniture, and dimensions and structure are hard to keep accurate. Pieter Levels's Interior AI is a well-known example, and he says there was little new technology to use in the past 12 months. Now that AI can generate 3D scenes from video or floor plans, these apps can move from rendering images to offering editable spaces.

Architecture Highlights and Mechanism

According to his posts, a user uploads a walkthrough video of a living room or a whole floor, a few photos, or a floor plan. The system generates a headless Blender 3D scene, which Interior AI turns into an editable space. Users can recolor walls, floors, ceilings and any object, move or delete furniture or add pieces from a furniture library, and get a floor plan. Rendering runs on headless Blender on a VPS he set up himself, not on a third-party 3D API. He credited @wesbos for much of the inspiration but did not say which model or reconstruction pipeline he uses. He called it "a quick prototype but it works".

Benchmarks and Measured Results

He has published no accuracy, speed or reconstruction quality figures, and there is no third-party testing; everything above comes from his own demo videos. The verifiable numbers are engagement: the idea post had about 53K views and 275 likes, the prototype post about 43K views, 233 likes and 214 bookmarks, and the launch post about 19K views, 140 likes and 105 replies. Several developers in the replies asked how long one room or a full floor takes to process; in the replies we read, he had not yet given a number.

Developer Guide and Getting Started

You can try it at interiorai.com, a paid product he says is now open to all users. For developers building something similar, the case offers a few lessons. First, run the slowest step, rendering, on a VPS you control with headless Blender, so you own the cost and the queue. Second, ship a minimal edit set first (empty, move, recolor) and add advanced features such as style switching later. Third, settle pricing before launch: people in the replies are already asking whether he will keep subscriptions or move to pay-per-use.