Grid overlay on an image, flattened into the file

Yes, unless your picture is over 3,000 px on the long edge or you put the labels in a margin. Files above that are scaled down to 3,000 px before the lines go on, and the panel says so with both numbers, because silently returning a smaller file than the one you handed over is the kind of surprise that ruins a print. Labels set to sit in a margin add a white strip on the top and left edges — between 20 and 64 px depending on the picture, and stated exactly. Labels set to sit inside the cells add nothing, so the output stays pixel for pixel the input.

Because pixels are whole and your frame usually is not divisible. A picture 2,560 px across in 12 columns wants 213.33 px a cell, which no raster can draw, so eight columns come out 213 px and four come out 214 — and the difference has to go somewhere. This tool rounds every boundary from its exact position rather than stepping along a rounded pitch, which spreads the leftover pixels through the grid instead of dumping all of them into the final column. The table under the preview names the sizes and how many cells have each, so nothing about it has to be discovered by measuring the result.

No — the lines are composited into the pixels, and that is the operation rather than a shortcoming. A flattened grid survives being pasted into a message, uploaded to a forum that strips metadata, or printed by somebody who has never opened a layered file, which is why people ask for this instead of doing it in an editor. The cost is that it cannot be undone: painting over a line does not recover what was underneath it. Keep the original and treat the ruled file as a copy. If you need the grid switchable, the right tool is an image editor with layers, not this page.

Yes, and the transparency is preserved. The picture is drawn first and the lines over it, so a transparent background stays transparent and the grid lines cross it as lines rather than as a filled sheet — useful when the ruled file is going over a colored background later. The one exception is the label margin: that strip is filled white so the numbers are readable, because a label drawn on transparency is invisible against half the backgrounds it might land on. Choose labels inside the cells instead if the file has to stay transparent to its edges.

Use the letters-down option and say B4, because it cannot be read in the wrong order. Numbers on both edges are quicker to scan but 2,4 and 4,2 are different squares and half the people in a conversation will assume the opposite convention from the other half — the row-then-column argument is old enough to have caused real errors in map reading. Letters down the side and numbers across the top removes the ambiguity at no cost, and the CSV export uses the same naming as the picture so the two can be read side by side.

No, it draws one file and hands you the coordinates to cut it yourself. Producing 144 numbered files raises questions this page cannot answer — what they are named, whether the edges overlap, what happens to the odd pixel — and getting any of them wrong quietly is worse than not doing it. The CSV lists every cell as a column, a row, and a left, top, width and height in pixels, which is what a crop script or a batch action in an editor consumes. The rectangles tile the frame exactly, so the pieces cut from them join with no seam and no gap.

Somewhere between 4 and 16 for anything a person will read, and far higher only when a machine is reading it. A grid of 8 by 8 on a 1,600 px picture gives 200 px cells, each large enough to hold a legible label inside it. Past about 40 divisions the cells fall below 40 px on a typical file, labels no longer fit inside them, and the lines start to be a meaningful fraction of what you can see — at which point the tool stops you: a cell narrower than four line widths is refused rather than drawn, with the thinner line offered as one click.