Raising the picture out of the surface, cutting it in, projecting it as a texture and letting it control material thickness are four separate jobs with different tools and different numbers. Here is what each one needs and how to check it before printing.
Imprinting an image on a 3D model covers four different operations: raising the picture above the surface, cutting it into the surface, painting it across the surface as a texture, and letting it control how much light passes through the part. They need different tools, they fail in different places, and only some survive the trip through a slicer intact.
This tutorial answers the questions that decide the result: whether the picture has to become geometry, how deep a relief should be for a 0.4 mm nozzle, which free tools do each job, and when generating a new object from the image beats decorating one you already have.
Two methods change the geometry: the picture becomes raised material or missing material. The other two leave the shape alone and change only the surface data or the material thickness. STL stores triangles and nothing else, so anything living in a texture vanishes at export. Pick the method from what the finished part has to do, not from what looks best in the viewport.
| Method | What it changes | Good for | Free tools | Starting numbers |
|---|---|---|---|---|
| Emboss (raised) | Adds height above the surface | Logos, monograms, plaques, shallow portraits | Blender Displace, Fusion 360 Emboss, Tinkercad | Relief 0.4-1.0 mm, 0.2 mm layers |
| Engrave (recessed) | Removes material from the surface | Text, dials, inlays, paint-filled lines | Blender boolean, FreeCAD, OpenSCAD | Depth 0.6-1.5 mm, stroke 0.8 mm |
| Colour projection | Nothing: it paints the texture | Miniatures, renders, game assets | Blender stencil paint, Substance Painter, Krita | 2048-4096 px maps, 8-16 px bleed |
| Lithophane | Thickness follows brightness | Portraits, landscapes, lampshades | itslitho, Lithophane Maker, Cura plugin | 0.8-3.2 mm thick, 0.1 mm layers |
| AI image-to-3D | Builds new geometry from the picture | One object with no model to decorate yet | Open3DAI, Hunyuan3D, TRELLIS 2, TripoSR | 1-4 views, 20k-200k triangles |
If the part will be printed in one colour of filament, the image must exist as geometry, because the slicer only sees the mesh. If it will be rendered, animated, or printed on a machine that mixes filaments, a projected texture is cheaper, sharper and faster to redo when someone sends a new logo. A texture also costs no print time.
Both methods read the same greyscale image, where white means high and black means low. An emboss adds that height map on top of an existing surface; an engrave subtracts it, or subtracts an extruded vector shape from a solid. On FDM, keep the relief between 0.4 mm and 1.0 mm on a decorative plate, and between 1.0 mm and 2.0 mm where the image must survive sanding. An 8-bit file gives 256 steps across the relief, far finer than any layer height, so 16-bit depth matters only for resin and CNC.
Height maps fail in three predictable ways: JPEG blocking turns flat areas into noise, hard edges turn into spikes that overshoot the relief depth, and a low-contrast photo produces a relief nobody can see. Convert to greyscale, raise contrast until the darkest and lightest tones are used, blur by one or two pixels, and keep the source at least 1024 px on the long edge. Thin lines should be redrawn as vector paths, not sharpened.
For logos and lettering, a boolean cut beats displacement: the walls stay vertical and the slicer has nothing to interpret. Extrude the SVG to the engraving depth, keep it a closed solid, subtract it from the base and delete the cutter. Keep strokes 0.8 mm or wider to fit two perimeters in the groove, and leave 1.5-2 mm of material under the deepest cut. Cap heights under about 4 mm are mush on FDM and clean on resin.
Projection paints a mesh that already exists: place the picture in front of the surface, paint or bake it through a camera or a stencil, and store the result in a UV texture. Blender does it with the Stencil brush in Texture Paint mode or by projecting from view and baking, and Substance Painter has a dedicated projection tool. Texel density decides how sharp the result looks: about 10 px per millimetre is a common game target, while a 32 mm miniature wants 20-30 px per mm, which pushes a 2048 px map to its limit. Leave 8-16 px of padding around UV islands, and note that a projection stretches wherever the surface turns away from the camera axis.
A lithophane prints the image as thickness: dark pixels become thick plastic, bright pixels stay thin, and a backlight turns the map back into a picture. Typical settings are 0.8 mm at the thinnest point, 2.8-3.2 mm at the thickest, 0.1 mm layers and 100% infill so the light path stays solid. Print it standing on its long edge, so the layer lines run across the image; add a brim, slow the outer walls to 30-40 mm/s, and use white or natural PLA. itslitho, Lithophane Maker and the Cura plugin all build the mesh and expose minimum and maximum thickness.
Sometimes the image is not decoration but the subject: a photo of a person, a pet or a prop that has to become an object. Image-to-3D generation builds a mesh from one picture, or from two to four views of the same subject, and bakes a texture onto it. Expect 20,000 to 200,000 triangles, with the front close to the photo and the back invented. That suits a figurine and fails anything that must fit an existing part, so re-project the photo in Blender if it has to stay sharp.
Two numbers decide whether the image survives printing. The first is layer height: a 0.3 mm relief printed at 0.2 mm layers has one or two steps to describe the whole picture, so keep the relief at least three or four layer heights tall, meaning 0.6-0.8 mm at 0.2 mm layers or 0.4 mm at 0.1 mm layers. The second is nozzle width: features narrower than 0.4-0.6 mm vanish or print as strings, and walls stay reliable in the 0.8-1.2 mm band. Reliefs print best lying flat, and colour inside the part comes from 3MF on a multi-filament machine or a filament swap at a chosen layer.
| Relief height or line width | Layer height | What you get |
|---|---|---|
| 0.2 mm | 0.1 mm | Two steps, best on resin |
| 0.4 mm | 0.1 mm | Readable on small parts, vertical walls |
| 0.8 mm | 0.2 mm | Four steps, the usual FDM choice |
| 1.5 mm and deeper | 0.2 mm | Bold relief, check the wall left underneath |
Open3DAI generates and exports meshes and textures. It does not retopologise, unwrap UVs, or promise print-ready output without cleanup, and the relief and lithophane mappings above are not part of it. Engine availability, credit pricing and third-party terms change, so check the vendor page and your slicer profile before committing a budget.
Direction of the same height map. An emboss adds material above the original surface, so the picture catches light and reads in a raking view; an engrave removes material, so the picture holds shadow, paint or a second filament. Engraving is the safer choice on thin parts, because a groove cannot collide with the geometry behind it, and embossing is the safer choice where the surface will be sanded or handled, because raised detail can be worn down.
Start at 0.4-0.6 mm and go up if the image still looks flat. Keep the relief at least three or four layer heights tall: 0.6-0.8 mm at 0.2 mm layers, or 0.4 mm at 0.1 mm layers. Decorative plaques often use 0.8-1.0 mm, and anything past roughly 2 mm stops reading as an image and starts reading as a separate object, while also collecting dust and print artefacts.
About 0.8 mm for a groove with two perimeters inside it, which is the width that survives sanding and paint. A 0.5 mm groove prints as a single extrusion and looks thin but works on small parts. Below roughly 0.4-0.6 mm the slicer either drops the line or prints it as a string. Lettering also needs height: about 4-5 mm cap height is the practical floor on FDM, while resin handles 1.5-2 mm.
For FDM, no. An 8-bit image gives 256 levels, which across a 0.8 mm relief is roughly 3 microns per level, two orders of magnitude finer than a 0.1 mm layer. Sixteen-bit files matter for resin printing, CNC relief carving and displacement work where the height range is large, and they help when you stretch contrast hard, because quantisation banding stays invisible.
Not with STL, which carries no colour at all. The routes that do work are 3MF on a multi-filament printer, a filament swap at a fixed layer height on a single-extruder machine, hand painting over a primed surface, or printing a lithophane and letting transmitted light create the image. Projected textures are for renders, game engines and AR; they vanish the moment geometry is exported to STL.
Roughly 0.8 mm at the brightest point up to 2.8-3.2 mm at the darkest, printed with 0.1 mm layers and 100% infill. Below about 0.6 mm the plastic becomes fragile and light leaks through unevenly; above roughly 3.5 mm the panel stops transmitting enough light to show the image. White or natural PLA is the usual material, and a frame around the panel protects the thin edge.
Yes, if the image is the subject. Image-to-3D generation turns one photo, or a set of two to four views, into a textured mesh, typically 20,000-200,000 triangles, which you then scale, repair and slice. The back and the hidden sides are invented rather than measured, so this suits figurines, display pieces and decoration, and it does not suit parts that must fit an existing assembly.
No. Open3DAI covers generation rather than surface editing: text-to-3D and image-to-3D from one image or a multiview set, multi-engine comparison, AI texturing, and export to GLB, glTF, OBJ, STL, PLY and 3MF. Relief mapping, boolean engraving, lithophane thickness curves and UV unwrapping happen in Blender, Fusion 360, Meshmixer or a dedicated lithophane tool, and no export is guaranteed watertight.
Send the same prompt or image to several engines, compare the meshes side by side, then export GLB, glTF, OBJ, STL, PLY or 3MF. Creating an account grants free credits and needs no card.
Silhouette extrusion, lithophane height maps and AI image-to-3D all end in the same file extension and almost nothing else. Here is what each route controls, which photos survive it, and the checks that happen before the first layer.
Texturing is four decisions taken in order: whether the model needs UVs, how the UV islands are laid out, which PBR maps you author, and where the pixels come from. Each decision is explained here with the numbers that make it.
One picture can become a screen-only model, a game character or a printed figure, and each route needs different work. Here is the five-step photo-to-STL pipeline, the print settings a small figure needs, and how to convert an existing AI action-figure image into a real object.