Here is the obvious way to get a photograph onto a sand table. Convert it to greyscale. Walk across it row by row. Where a pixel is dark, wiggle the ball; where it is light, move straight. Repeat for every row.
This is roughly how an inkjet printer works, it is straightforward to implement, and several sand table converters do exactly it. The results are consistently disappointing, and the reason is worth understanding — it explains what SandPath does instead.
Why printers get away with it
An inkjet lays down opaque dots at a few hundred per inch. Adjacent rows overlap slightly, ink bleeds a fraction into the paper fibres, and the eye integrates everything above a few dozen rows per inch into continuous tone. The row structure exists but is invisible.
None of those conditions hold in sand.
Why sand does not
The ball is enormous. A typical sand table ball is 10–15 mm across. On a 400 mm bed that is one row per 2.5% of the table's width — call it forty rows, total, versus the several thousand a printer would use across the same distance.
Rows cannot overlap. Two adjacent grooves do not blend; the sand between them stands up as a ridge. Where a printer gets smooth tone, sand gets corduroy.
Nothing lifts. A printer's head moves without printing between rows. The ball cannot. Every return sweep ploughs a groove of its own, so the "blank" parts of the image are as carved as the dark parts.
Sand has no tone. A groove is a groove. There is no darker groove for a darker pixel — you can only vary spacing or amplitude, and both are limited by the ball's width.
Put together: scanlining a photograph gives you forty parallel furrows across the table with some wiggle in them. From across a room it may resolve into something. Up close, which is where people actually look at sand tables, it is visibly a raster.
What sand is actually good at
The medium's strength is the opposite of a printer's. A sand table draws one continuous, elegant line of arbitrary length and curvature. The whole aesthetic — the reason people buy these things — is a single unbroken path wandering through the sand.
So the right conversion strategy is not "reproduce the image." It is find the lines in the image and draw those. That is a vectorisation problem, and it has three good answers depending on what the source material is.
Outline: edge detection
Outline mode runs Sobel edge detection, convolving the image with a pair of gradient kernels — one horizontal, one vertical — to find where brightness changes sharply. Combining their responses gives edge strength at every pixel; thresholding and following those edges gives contours.
The output is a sketch: the boundaries in a photograph, without any filled areas. This is the mode for photographs and detailed artwork, and the mental model to hold is a line drawing done by someone tracing the photo, not a reproduction of it.
It works best when the subject is distinct from the background. Busy scenes produce edge detections everywhere and the result reads as noise.
Threshold: marching squares
Threshold mode splits the image into two regions at a brightness cutoff, then extracts the boundary between them using marching squares — an algorithm that examines each 2×2 cell of the binary grid and emits the contour segment passing through it, then stitches the segments into closed loops.
Because the output is closed, well-defined outlines, this is the strongest mode for logos, silhouettes, lettering, and high-contrast graphics. It is also the mode most sensitive to its threshold: a photograph run through it typically yields either a black blob or nothing, depending on which side of the cutoff the image sits.
Centerline: skeletonisation
Centerline mode iteratively erodes shapes until only a one-pixel-wide medial line remains, then traces that.
The distinction from Outline matters most on line art. Trace a drawn stroke with edge detection and you get two lines — one down each side — which in sand reads as a hollow tube. Skeletonise it and you get one line down the middle, which is what the original stroke was. For handwriting, sketches, and pen drawings this is the mode that behaves the way you expect.
After tracing: simplification
Whichever mode runs, the traced paths carry far more points than the ball can express — often one per pixel. Douglas–Peucker simplification recursively removes points lying within a tolerance of the line between their neighbours.
The effect is dramatic and nearly free: point counts commonly drop by an order of magnitude while the visible shape is unchanged, because most of the discarded points described detail finer than the ball's diameter. That is the Path Detail control.
Choosing well
The single highest-leverage decision is the trace mode, and the failure mode is predictable: photographs run through Threshold, line art run through Outline. If a conversion looks wrong, change the mode before touching anything else.
And accept the medium. A sand table renders an interpretation of a photograph — its edges, its structure — not the photograph. Sources that survive the trip are ones whose meaning lives in shape rather than tone.
Ready to try it? Open the converter, or read how the conversion works.