How to Turn an Image into a 3D Model: 4 Methods Compared

One flat image does not contain the depth of the object it shows, so every image to 3D method either adds the missing depth by rule, by measuring more photos, or by guessing.

Updated 20 October 2026 · 7 min read

Searching for how to turn an image into a 3D model brings up very different tools that all claim to do the same thing. They do not. A photo is a 2D grid of colours; it has no information about the back of the object or how far away each point is. Every method fills that gap in a different way, and the right choice depends on whether you want a flat relief, a crisp logo, an accurate copy of a real object, or a rough approximation of something you only have one picture of. Here is what each approach actually produces and when to use it for 3D printing.

Quick comparison

Method 1: Heightmap relief and lithophanes

A heightmap converts brightness into height. Each pixel becomes a point on a surface, and its grey value sets how far that point rises above the base. Dark can mean thick or thin depending on the setting. The result is sometimes called 2.5D: there is real depth, but only in one direction, and nothing behind the surface.

Two common uses:

• Lithophanes – a thin plate where thick areas block light and thin areas let it through. Held against a lamp, the plate shows the photo. Here dark pixels map to thick plastic.
• Reliefs and embossing – the surface is meant to be seen in reflected light, like a coin or a plaque. Light pixels are usually mapped high.

Heightmaps work because the conversion is exact and predictable. Their limitation is that brightness is not depth: a white shirt becomes a tall plateau and a black shadow a deep valley, even though neither matches the real shape. For lithophanes this does not matter, because brightness is what you want to reproduce. For a "3D portrait" seen from the side, it looks wrong.

Tools: the image to STL converter builds a relief from any PNG or JPG, with control over size, base and maximum height. The lithophane maker adds the settings that matter for backlit prints, such as minimum and maximum thickness, frames and curved shapes.

Tips for good reliefs

• Increase contrast and remove busy backgrounds before converting.
• Blur noise slightly; JPEG artefacts become visible bumps.
• Match resolution to the printer: with a 0.4 mm nozzle, detail finer than about 0.2–0.3 mm per pixel cannot be reproduced, so a 100 mm wide plate gains nothing above roughly 500 pixels across for a relief. Lithophanes benefit from a bit more because layer height and wall thickness carry the detail.

Method 2: Logo and SVG extrusion

If your image is a logo, icon, letter or line drawing, you do not want a heightmap. You want the outline traced and pushed straight up to a fixed depth, giving clean vertical walls and flat tops.

The ideal input is an SVG vector file, because it stores exact shapes instead of pixels. The SVG to STL tool reads the paths, extrudes them to the depth you set and keeps holes in letters like O and A correctly open.

If you only have a raster image (PNG, JPG), you first trace it into vectors. This works well for high-contrast, flat-colour artwork and poorly for photos. Practical steps:

1. Start from the largest, cleanest version of the image you can find.
2. Reduce it to pure black and white with a threshold adjustment.
3. Trace it to SVG in a vector editor (Inkscape's Trace Bitmap is free) or with an online tracer.
4. Delete stray specks and simplify very jagged curves.
5. Extrude to STL and check the thinnest feature is at least two nozzle widths (0.8 mm on a 0.4 mm nozzle).

For plain text, skip the image entirely: text to STL builds letters directly from fonts, which is cleaner than tracing a screenshot.

Method 3: Photogrammetry from many photos

Photogrammetry reconstructs real geometry by matching features across many photos taken from different positions. Because each point is seen from several angles, the software can triangulate its actual position. This is the only method here that measures the object rather than inventing depth.

What you need

• Photos: typically 30–100 for a small object, more for large or detailed ones, with roughly 60–80% overlap between neighbouring shots.
• Coverage: circle the object at two or three heights, then add top views. Any area not photographed becomes a hole.
• Lighting: even and diffuse. Hard shadows and reflections confuse feature matching.
• Surface: matte and textured. Shiny, transparent and plain single-colour objects reconstruct badly; a light dusting of scanning spray or dry shampoo helps.
• Software: options include Meshroom (free, open source), Apple's Object Capture via apps on Mac and iPhone, and phone apps such as Polycam or KIRI Engine. Some phones also use their LiDAR sensor, which is useful for rooms but too coarse for small parts.

Preparing a scan for printing

Scans come out with holes, floating debris, a chunk of table and often several million triangles. Crop and close the mesh in the scanning app or Blender, run it through STL repair to make it watertight, and reduce the triangle count to a few hundred thousand before slicing. Scale is also arbitrary unless you included a reference object of known size, so measure one feature on the real object and scale to match.

Method 4: Single-image generators

A newer category of tools takes one picture and returns a complete 3D mesh, usually as GLB or OBJ with a texture. They are trained on large collections of 3D models and predict a plausible shape, including the sides the photo does not show.

What they do well

• Characters, toys, creatures and stylised objects from a clear, front-facing image.
• Quick concept models and game or AR props where texture carries the look.
• Objects with simple, symmetrical forms.

Where they fall short

• Hidden sides are invented. The back of a product or a person's profile is a guess, not a measurement.
• No dimensional accuracy. Do not expect a replacement part, bracket or anything that must fit.
• Thin features such as fingers, straps and text are often merged, missing or too thin to print.
• Colour lives in the texture, which an FDM single-colour print ignores. Details that looked sharp on screen can turn out to be painted on rather than modelled.
• Mesh quality varies: outputs may be non-manifold, have internal faces or very high triangle counts.

Before printing a generated model, open it in the GLB viewer and switch off textures if you can, so you see the bare geometry the printer will produce. Then convert, repair and reduce it like a scan. Check the terms of the generator you use, as commercial-use rights differ between services.

Which method should you use for 3D printing?

• A photo you want to display with light: lithophane.
• A photo as a decorative plaque or texture: heightmap relief.
• A logo, name, sign, stamp or cookie cutter: SVG extrusion.
• A real object you can walk around and photograph: photogrammetry.
• A character or object you only have one image of, where accuracy is not critical: a single-image generator, followed by repair.
• A functional part that must fit something: none of the above. Measure it and model it in CAD.

What our tools do and do not do

The browser tools on this site handle the first two methods directly and the clean-up after the other two:

• Do: convert images to heightmap reliefs and lithophanes (flat, curved and cylindrical), extrude SVG artwork and text to STL, view GLB/OBJ/PLY/STL output from scanners and generators, repair meshes, reduce triangle counts and scale models.
• Do not: reconstruct a model from multiple photos or generate a full 3D object from a single image. Use dedicated photogrammetry software or a generator for that step, then bring the mesh back for checking and repair.

FAQ

Can I turn a single photo into an accurate 3D model?

Not accurately. One photo does not contain depth or the hidden sides. You can make an exact relief or lithophane from it, or a plausible full model with a generator, but true geometry requires multiple photos (photogrammetry), a 3D scanner or CAD modelling.

What is the easiest way to make a 3D print from a picture?

A lithophane or relief. Upload the image, set the size and thickness, and download an STL. It works with any photo and prints without supports.

How do I turn a logo into an STL?

Use an SVG version of the logo if possible and extrude it with an SVG to STL tool. If you only have a PNG, convert it to black and white, trace it to SVG, clean up the paths, then extrude.

Are images converted by these tools uploaded to a server?

The image to STL, SVG to STL and lithophane tools here run in your browser, so the file is processed on your own device.