Image to height map conversion with a controlled range

Depth describes distance from a camera. Height describes elevation from a surface. Converting one into the other is an editorial act: you decide that the nearest object becomes the highest point, that the background becomes a flat plane, and that the intervals between them are distributed in a particular way. The conversion keeps the ordering the estimator found, and replaces everything else with your intent.

DepthCarve starts from camera-relative depth and then gives you the range controls that make the reinterpretation explicit. Inversion decides which end is raised, the black and white points decide how much of the histogram survives, and flattening removes the background so it does not consume half the available height.

For a PBR or displacement workflow you want the grayscale field itself: a single channel that a shader or a sculpting tool will multiply by its own height value. Precision matters more than physical scale, because the receiving tool defines the units. A true 16-bit PNG is the right container when the surface will be subdivided finely enough to show the difference.

For a physical relief you want a closed mesh with a stated width, height, and base. That path goes through the STL exporter, which resamples the same field to a printable grid and adds walls and a bottom. Keeping both outputs from the same approved field is what makes a design reviewable: the render and the carve then differ only by the machine, not by the image.

This is not a GIS or terrain generator. The values come from one photograph, carry no geographic coordinates, and have no automatic metric scale. A hill in the picture is only closer or farther from the camera; it is not higher in any absolute sense.

If you need real elevation, use survey or elevation data. If you need a plausible surface derived from a picture, that is exactly what this does, and the preview is the cheapest way to decide whether the picture contains enough structure to justify the work.