The 3D Handbook
Topic

Photo-sourced and scanned materials

Making materials from photographs, from a single photo converted to PBR maps, through multi-light photo sets, to full photogrammetry scans; de-lighting, cross-polarisation, Substance 3D Sampler, and scan libraries.

texturingphotogrammetryscanningmaterialssampler

A photo-sourced material starts from photographs of a real surface rather than from painting or nodes. The range runs from a single phone photo turned into physically based rendering maps in seconds, to hundreds of carefully lit photos reconstructed into a measured surface with photogrammetry. More photos and more control give more accurate height and colour, at the cost of time and equipment.

Photo-sourced materials end up used like any other: as tileables, decals, trims or unique textures. See ways to texture an environment.

#Three levels of capture

Three panels. Left, single photo: one camera above a surface, one photo, software guesses height and removes shading. Middle, photo set with moving light: a fixed camera above the surface and a light shown at several angles around it, 2 to 8 photos. Right, photogrammetry: many camera positions in an even grid over the surface, around 300 photos per 2 square metres, reconstructing real geometry.
Three levels of capture. The single photo infers height; the multi-light set measures how light falls; photogrammetry reconstructs the real surface.Diagram by the handbook author

#1. Single photo to material

Software estimates height, normal and roughness from one image and tries to remove the lighting baked into it. Adobe's Image to Material in Substance 3D Sampler "allows to generate a high-quality PBR material from a single input image" with two algorithms (Substance 3D Sampler documentation: Image To Material):

  • AI Powered uses machine learning trained on "fabrics, organics, indoors and outdoor surfaces" to create normal, height and roughness and to remove shadows and highlights from the albedo.
  • B2M (bitmap to material) is procedural; it "may produce less accurate results but will work on a wider range of input images".

The result is an estimate. It works well for fairly flat surfaces such as fabric, plaster or paving, and poorly for deep forms such as rocky ground, where the software cannot know what is hidden.

#2. Photo sets under different lights

Keep the camera still and move the light. Each photo shows how the surface responds to light from one direction, which gives much better normals than a single image. Sampler's Multiangle to Material "creates a material from 2 to 8 input images taken under specific light conditions", for example eight photos lit at 0°, 45° and so on to 315°, and outputs base colour, normal, height, metallic and roughness (Substance 3D Sampler documentation: Multiangle To Material). Adobe notes that such lighting can be achieved with a material scanner (Substance 3D Sampler documentation: Multiangle To Material).

#3. Photogrammetry scans

Photograph the surface from many overlapping positions and let software reconstruct real geometry. Poly Haven's published process (Poly Haven Blog: Photoscanned Texture Creation Process):

  • Capture "around 300 images per 2 metre square area", spaced as evenly as possible, with measuring tapes for spacing and scale and a colour chart on the same plane as the surface.
  • Process RAW files in batch (brightness, white balance, vignetting), avoiding per-pixel edits such as sharpening.
  • Reconstruct in RealityCapture: align, build a dense mesh, simplify, unwrap and project colour.
  • Bake height, normal and colour onto a flat plane in Blender.
  • Make it tile by removing gradients and seams and fixing artefacts.

The core outputs are only three maps, colour (albedo), displacement and normal, because "all the other maps can be derived from" them, saved as "a 16 bit lossless format such as PNG" (Poly Haven Blog: Photoscanned Texture Creation Process).

See photogrammetry for capturing whole objects and photogrammetry and scanning tools for software.

A photogrammetry scan captured in neutral lighting, delivered with albedo, normal, roughness and displacement maps of up to 8K (Sketchfab: Old Village Building Ruins Scans). This is the lighter version at 181,093 faces. Orbit close to the walls: the stone colour and the small relief come from the scanned maps, while the large broken shapes are real reconstructed geometry.Pers Scans, via Sketchfab

#De-lighting

A photo records the surface and the light that fell on it. A base colour map must contain only the surface. Removing shadows, highlights and gradients is called de-lighting.

You can do it at capture time, in processing, or both:

#Cross-polarisation

Cross-polarisation removes reflections at capture. A linear polarising filter sits over the light, and a second one on the lens is rotated 90 degrees to it. Paul Bourke explains why this works: "shiny surfaces (metals, specular materials) preserve polarisation of the light they reflect", while diffuse surfaces "randomly depolarise" it. So specular reflections are "greatly attenuated" while diffuse reflections still pass (paulbourke.net: Cross polarisation in photography). What reaches the camera is close to pure surface colour. The cost is light: Bourke measured a loss of at least 1.3 stops, usually made up with a brighter source (paulbourke.net: Cross polarisation in photography).

To line the filters up, Bourke photographs a metal spoon in live view and rotates the lens filter until its highlight disappears (paulbourke.net: Cross polarisation in photography).

Poly Haven calls "a cross-polarized flash setup" ideal "for all surfaces to avoid reflections" (Poly Haven Blog: Photoscanned Texture Creation Process). They also recommend taking extra reference photos, especially one showing the surface's specular properties, because cross-polarised photos no longer show them (Poly Haven Blog: Photoscanned Texture Creation Process). You need that reference to set roughness by eye later.

A camera with a lens filter and a flash with a filter rotated 90 degrees above a surface. Two light paths: the specular reflection bounces straight off the surface and is blocked at the lens filter; the diffuse light scatters inside the surface and passes through the lens filter into the camera.
Crossed filters block mirror-like reflections but let diffuse colour through, which gives a cleaner base colour.Diagram by the handbook author

#Scan libraries

You do not always need to scan yourself:

  • Poly Haven: textures, HDRIs and models, "all licensed as CC0", usable commercially without attribution (polyhaven.com: License).
  • Megascans moved to Epic's Fab marketplace. It was made free to all until the end of 2024, and Epic announced that from 2025 most Megascans content would be paid, with individual assets priced from $0.99 (CG Channel: Epic Games has made Megascans free to all…).

See asset libraries and asset licences before using any library asset in a commercial project.

#How to make one

These steps turn one photo of paving into a tiling material in Substance 3D Sampler. For a measured scan of the same surface, follow Poly Haven's process above instead.

  1. Shoot the photo. Photograph a flat patch of paving straight on, in overcast light or shade (Poly Haven Blog: Photoscanned Texture Creation Process), with a tape measure in frame. Aim for a patch about 1 × 1 m that fills the frame, and shoot RAW if you can, because 16-bit sources band less when you correct them (Poly Haven Blog: Photoscanned Texture Creation Process).
  2. Prepare the image. Correct white balance and vignetting and crop to the 1 × 1 m area. Avoid sharpening and other per-pixel edits (Poly Haven Blog: Photoscanned Texture Creation Process).
  3. Create the material. In Sampler, drop the image in and apply Image to Material. Try AI Powered first; switch to B2M if the subject is unusual, since it "will work on a wider range of input images" (Substance 3D Sampler documentation: Image To Material).
  4. Check the de-lighting. Look at the base colour on its own. Remaining shadows or gradients mean more correction, or a better photo (Substance 3D Sampler documentation: Image To Material).
  5. Make it tile. Use Sampler's tiling tools to remove seams, then preview the material 3 × 3 and fix any stone that forms an obvious repeat.
  6. Set roughness by reference. Compare roughness against a photo that shows the surface's shine, because cross-polarised or flat-lit photos do not show it (Poly Haven Blog: Photoscanned Texture Creation Process).
  7. Export at a known size. Export base colour, normal, height and roughness at 2048 px and note that the texture covers 1 × 1 m, which gives 2048 px per metre. See texel density.

#Video tutorials

#Use in recent games

#Further reading and watching

References

  1. Adobe · Image To Material · Substance 3D Sampler documentation · docs · accessed 2026-10-21
  2. Adobe · Multiangle To Material · Substance 3D Sampler documentation · docs · accessed 2026-10-21
  3. Rico Cilliers (with comments by Greg Zaal) · Photoscanned Texture Creation Process · Poly Haven Blog, 2022 · article · accessed 2026-10-21
  4. Pers Scans · Old Village Building Ruins Scans · Sketchfab · web · accessed 2026-10-22
  5. Adobe · 3D Capture: Equipment · Substance 3D Sampler documentation · docs · accessed 2026-10-21
  6. Paul Bourke · Cross polarisation in photography · paulbourke.net, 2022 · article · accessed 2026-10-21
  7. Poly Haven · License · polyhaven.com · web · accessed 2026-10-21
  8. CG Channel · Epic Games has made Megascans free to all, but only until the end of 2024 · CG Channel, 2024 · article · accessed 2026-10-21
  9. Tsvetelina Valkanova (Ubisoft Sofia) · A Look at the Material Workflow in Assassin's Creed Valhalla · 80 Level · article · accessed 2026-10-21
  10. Yulong Huang (interviewed by Theodore McKenzie) · How to Make Tool Kit With History Using ZBrush & Substance 3D Painter · 80 Level, 2023 · article · accessed 2026-10-21
  11. David Hodgson · Modern Warfare Initial Intel: How Photogrammetry is Helping to Shape Call of Duty: Modern Warfare into a new high watermark for graphics in gaming · Activision Blog, 2019 · article · accessed 2026-10-22
  12. Roland Austinat · The Making of Star Wars: Battlefront: 100.000 Fotos für vier Planeten · PC Games Hardware, 2016 · article · accessed 2026-10-21

Last updated 2026-10-22