Made by the FurniMesh team. FurniMesh (furnimesh.com) is our own product, and we recommend it where it fits.

Guide

Relief vs full 3D vs photogrammetry

A relief, an AI-generated model and a photogrammetry scan can all start from pictures, but they build very different things. This comparison shows what each method can and cannot give you.

There are three ways to get 3D out of pictures, and each answers a different question. A relief asks how bright each pixel is. An AI generator asks what the object probably looks like all the way round. Photogrammetry asks where each surface point really is, measured from many photos. FurniMesh (furnimesh.com), our own product, is an AI generator for furniture; the relief maker on this site makes reliefs. This guide compares all three methods so you can pick the right one.

A stone bas-relief panel showing a skeletal figure and a muscular figure on either side of a caduceus
A sculpted bas-relief. Digital reliefs work the same way: shapes rise from a flat background, with no back or sides. Marc Merlin · CC BY-SA 4.0 · via Wikimedia Commons · file page

The three methods at a glance

Relief (height map)AI full modelPhotogrammetry
InputOne pictureOne picture (sometimes a few)Dozens to hundreds of overlapping photos
What it buildsA raised panel on a flat baseA complete object with inferred hidden sidesA measured surface of what was photographed
Back and sidesNone (flat base)Predicted by AIReal, if you photographed them
TextureThe picture itself, on topBaked from the photoBaked from all photos
AccuracyNot a shape measurementPlausible, not measuredMeasured; scale needs a reference
EffortSecondsMinutesA capture session plus processing
Runs whereCan run in your browserOn the vendor's servers or a strong GPUPhone app, cloud or desktop
Best forSigns, plaques, lithophanes, texturesFurniture, products, props, charactersReal objects, sculptures, terrain, heritage

Relief: brightness becomes height

A relief reads each pixel's brightness and raises the surface by that amount. White becomes the highest point, black the lowest, or the other way round for lithophane-style panels lit from behind. Add a solid base and side walls and you have a closed shape you can print. That is exactly what the relief maker on our home page does, in your browser.

The method is fast, private and predictable, but it does not understand the picture. A white shirt and a white wall rise to the same height; a dark eye socket sinks into the slab. That is why reliefs look best with pictures where brightness already follows shape: carvings, embossed metal, side-lit faces and high-contrast logos.

Our tests show the numbers involved. A 300 by 200 pixel black-to-white gradient, made into a 120 mm wide relief with a 2 mm base and 5 mm of depth, produced a closed mesh of 30,990 triangles, 120 by 80 mm, rising from 2 to 7 mm, with a volume of 43.2 cubic centimeters, matching the expected 120 x 80 x 4.5 mm average almost exactly.

AI full model: the shape is predicted

An AI image-to-3D generator has learned, from large amounts of 3D data, what objects tend to look like. Given one photo of a chair, it segments the chair, estimates its depth and predicts the back, the far legs and the underside. Then it bakes the photo's colors onto the new surface.

The result is a complete object you can rotate, light and place in a scene, which is what most people mean by "picture to 3D model". Its limit is that hidden parts are predictions. They are usually plausible and sometimes wrong in detail, so check the model before production use. Specialized tools do best on their own category: FurniMesh on furniture and single objects, general tools such as Meshy and Tripo on characters and props.

Photogrammetry: the shape is measured

A rocky climbing crag with a fence in front, one frame from a photogrammetry dataset
One frame from a published photogrammetry dataset of a climbing rock. Many overlapping shots like this are combined into one measured model. Zbytovsky · CC0 · via Wikimedia Commons · file page

Photogrammetry finds the same features in many overlapping photos and triangulates their positions in space. With enough coverage, it reconstructs the real surface: every bump in a rock, every chip in a sculpture. Nothing is guessed, so nothing that was not photographed appears in the model.

That makes it the method of choice for heritage scanning, terrain and real objects you need to reproduce faithfully. It needs the object in front of you, consistent light, a lot of photos and some practice. Shiny, transparent and plain untextured surfaces are hard, because the software cannot find matching features on them. Polycam's free plan, with 10 captures of up to 150 images each, is an easy way to try it.

Which method should you use?

  • You have one photo of a furniture piece or product and want a model to place in a design or show in AR: an AI full model.
  • You have the object and need its real shape: photogrammetry.
  • You want a printed sign, plaque, coin, tile or lithophane: a relief.
  • You have a photo of a room: none of these turns it into a usable room model. Use a room planner for layout, and generate the individual pieces separately.

The methods also combine well. A furniture designer might generate a sofa with an AI tool for a room visualization, scan a carved detail with photogrammetry, and print a relief of the company logo for the showroom, each with the method that suits it.

Printing each kind of result

Reliefs are the easiest to print: the relief maker's STL is closed, flat-bottomed and sized in millimeters. AI models and photogrammetry scans usually arrive as GLB or OBJ. Convert them to STL, check the scale, since many tools do not output real-world size, and look for holes and paper-thin parts before slicing. Photogrammetry scans also often need their base cut flat and the hole underneath filled, because the underside could not be photographed.