The Furniture Industry Is Replacing Product Photography With 3D Files

A small workshop with six staff and forty chairs and stools found a new field on a marketplace listing form this spring: upload a 3D model. Not strictly required, but required for featured placement, and the same file now drives the rotating viewer on the shop’s own product pages, the “view in your room” button on the marketplace and the AR preview on an iPhone. The workshop had eight weeks and no 3D files at all. This is how it got forty of them, what each route cost, and where the studio photographer still fits, which is not nowhere.

One file, five places

The GLB format, a single binary file carrying the mesh and its textures, has become the common currency. A shop’s product page shows it as a rotating model with an AR button. The marketplace uses the same file for the room preview. Shopping results in search can display it through Scene Viewer on Android phones. iPhones and iPads want a second file in USDZ for Quick Look, which is a conversion rather than a new model. Some retailer room planners also accept FBX. The point is that one decent model, made once, is reused everywhere a customer might look at the chair.

The listing form was specific about what decent means. GLB and USDZ. Under 100,000 triangles. Textures at 2,048 pixels square. The model at real size in meters, origin at the center of the base, Y axis up. Under 15 MB. And a hidden side that is modelled with no holes, since the viewer lets the customer spin the chair round. Those eight lines decided the routes, because a file that fails any of them is rejected on upload.

Route one: a render from the drawings

Twenty-eight of the forty chairs had CAD drawings from the maker, which made them the easy cases. For those a freelance 3D artist turned each drawing into a model in about three hours at 180 dollars a chair, using fabric scans the workshop supplied at 300 dpi, so that the hidden side was modelled rather than guessed and the fabric was exact down to the weave, which matters when the fabric is the thing being sold. The workshop’s part of the job was small but not optional: the DWG, two fabric scans, two reference photographs and the width written on the drawing, followed by an approval of each finished model against a photograph of the real chair, which caught one chair with its arms 30 mm too low. The renders ran at seven a week from week two to week five.

Route two: a scanner for the carved pieces

Four of the chairs were a different problem. They are carved and upholstered, the drawings for them date from the previous owner and were not worth the paper, and a renderer working from a photograph would have had to invent the relief. They went to a scanning bureau instead, a structured-light rig at 250 dollars a chair, ninety minutes each plus cleanup. The owner grumbled about the price until he saw what a renderer had made of the carving on a test chair, at which point the grumbling stopped. One day of scanning in week three covered all four.

Route three: a photo for the plain ones

That left eight plain stools with neither drawings nor carving, and nobody who makes stools will pay 180 dollars for a render of one. The stools were photographed by a member of staff against the white workshop wall, from seat height, and each photograph went into FurniMesh in a browser, where the rotating preview was checked, the GLB downloaded and the seat width typed in to set the scale. Ten minutes a stool, one afternoon in week two, on a monthly plan shared across the batch.

A photo-derived model’s hidden side is inferred by the software from the visible side, and on a plain stool the inference is indistinguishable from the real thing. A carved chair is another story. The software would have smoothed over the carving with a polite guess, which is why those four went to the scanner and the stools didn’t. By the second stool somebody had written the photo rules on the wall above the bench in marker pen, and they stayed there for the rest of the month. One object. White wall. Camera at seat height. Three-quarter angle. At least 1,500 pixels on the short side. Flash off. And the seat width on a sticky note before anyone pressed upload, which is the rule every photo-to-3D tool publishes and the one everybody forgets.

The checks before upload, and the four that bounced

Every 3D file was opened in Windows 3D Viewer or macOS Preview, measured against the tape, counted in Blender’s statistics overlay, checked for embedded textures, converted to USDZ in Apple’s Reality Converter, checked for size and checked for sitting on the ground plane rather than floating above it. Week six, all forty.

Four still bounced on the first upload in week seven. A lounge chair came in at 118,000 triangles and was decimated to 60,000. A USDZ came in at 19 MB and was fixed by resizing its textures to 1,024 pixels. A stool floated 40 mm above the ground, fixed by moving its origin to the base. And one chair arrived with its textures missing, re-exported with them embedded. Every rejection was one of the eight lines on the form, which is the argument for reading the form first.

The 3D file naming was the dull part that saved the most time. sku_fabric_v1.glb, so dc01_oatmeal_v1.glb for the first dining chair in oatmeal, with the USDZ under the same name, one folder per product and the version bumped on any change. Forty products with up to six fabrics each is a lot of files to lose.

What a 3D file costs against a shoot day

The forty files came in under 6,500 dollars, which is 5,040 for the twenty-eight renders, 1,000 for the four scans and one month of a plan for the eight stools. For comparison, a studio day at this workshop runs to 1,800 dollars and gets through about ten chairs at five angles each, so the same forty chairs photographed on white would have been four days and 7,200 dollars for pictures that show each chair in exactly five positions and no others.

Where the comparison stops being close is the next fabric. A new colour on a photographed chair means booking the studio again, whereas on a 3D file it is a texture swap of about twenty minutes, and the dining chair that comes in four fabrics now has eight files where it once needed twenty photographs. The lounge chair in six fabrics is twelve files. The dimensions on the product page are read from the model instead of typed into a form, which ended a long-running argument about whether a particular stool was 450 or 460 mm wide. It was 455.

Lifestyle shots are still needed, one per range, and the photographer still does those. What the 3D file replaced is the product shot on white, five angles per chair, repeated every time a fabric changed. The studio day moved from a cost of goods to a marketing spend, and the photographer’s calendar got lighter without emptying.

What changed on the 3D file listings

Before the eight weeks: no 3D models, no AR, fabric variants shown only where a photograph existed, dimensions typed by hand. After: forty of forty with a model and an AR view, every fabric variant shown, and the dimensions read from the file. The week eight check was three phones on the workshop floor, an iPhone with Quick Look, an Android with Scene Viewer and an older Android that only managed the rotating model. All three put a chair at real size on the floor, which is the moment the point of the whole exercise becomes obvious to everyone watching.

Read Next

A few related pieces worth your time:

• Running Freight Taught Us About Building a Transport Management System
• 17 Print Server Alternatives for Enterprise IT
• 4 Unusual Ways to Protect Field Tech Gear

For a maker starting this month with 3D Files

Read the listing form before anything else and write down its limits. Measure every product once, one width each, and keep the list. Sort the products into three piles by what already exists: drawings to a renderer, carved pieces to a scanner, plain pieces to a photo tool. Take all the photos on one day against one wall. Run the checks on every file before uploading, since the marketplace will run them anyway. And test the AR on a phone before the first customer does.