Product

# Product Photo Previews Beat 3D Models You Do Not Have

Retailer AR only previews what someone modelled in 3D. A photo, a background cut-out, and real dimensions cover the second-hand cabinet too.

CodeLoop Studio·August 29, 2026·5 min read

![Ceramic vase photographed as a product photo while shopping](https://www.codeloop.ba/_next/image?url=%2Fimages%2Froomfit%2Fceramic-vase-product-photo.jpg&w=3840&q=75)

Retailer AR previews the catalogue someone paid to model in 3D. That is a small slice of what people actually put in their homes: the marketplace cabinet, the lamp from a shop with no website, the table a local carpenter made last month.

In [RoomFit](https://www.codeloop.ba/study-case/roomfit) the preview starts from a photo instead. Four steps: take the product photo, isolate the item from its background, enter the real width, height, and depth, then place it in the room.

## A visual approximation, and we say so

What you get is the item's appearance held at true dimensions — not a reconstructed 3D model you can walk behind. Calling it a 3D model would be the easy marketing line and the reason for every one-star review that follows.

The part that stays exact is the size, because that comes from numbers you typed rather than from the photo: [true-size AR with no pinch-to-scale](https://www.codeloop.ba/blog/true-size-ar-furniture-preview).

## What made the cut-out hard

* **Shop photos are not studio photos.** Real input is a phone shot under mixed lighting with three other sofas behind the one you want.
* **Edges cost more than masks.** A roughly correct silhouette with a clean boundary reads as believable; a precise mask with halo fringing reads as broken.
* **Transparency has to survive the pipeline.** Every resize, cache, and composite step is somewhere alpha can be flattened to white.
* **Give the user a retake, not an apology.** When isolation fails, the fastest recovery is a hint about a plainer background and a second shot.

## The same photo powers Match

Once an item is isolated, its colour, warmth, contrast, saturation, and pattern intensity can be compared against a saved room — that is RoomFit Match, a shopping aid rather than a verdict on your taste. The basic comparison runs locally; [cloud AI only runs when the user asks for it](https://www.codeloop.ba/blog/on-device-image-analysis-privacy), and those deeper checks are [metered credits](https://www.codeloop.ba/blog/ai-credits-in-consumer-apps).

## If you are building something similar

Image pipelines look trivial in a demo and eat weeks in production. We build this kind of work as [custom software](https://www.codeloop.ba/custom-software-development) and [mobile apps](https://www.codeloop.ba/custom-mobile-app-development) from Sarajevo. Scope it with us: [online estimate](https://www.codeloop.ba/online-estimate) or the [RoomFit case study](https://www.codeloop.ba/study-case/roomfit) first.