Writing the rail has no room for
Every number in this tool was argued before it was typed, and almost none of that argument fits beside a slider. A ceiling at eighty per cent. A default at twenty-five. A control that is simply not there. Each of those is a position, and a position stated without its reasoning is only an assertion with a house typeface.
So the reasoning lives here instead, at whatever length it takes. None of the three is a tutorial and none of them will teach you the interface — the rail is labelled, and if it needed an essay to operate it would be the wrong rail. What they are for is the moment a control surprises you: the smoothing did something you did not expect, the compare view showed more movement than you thought you had made, or you went looking for a slider that turned out not to exist.
Take them in any order. They were written to be read separately, which is why each one restates the small amount of the other two it depends on.
- What frequency separation actually splitsTwo bands, the radius that divides them, and the reason a Gaussian takes the one you meant to keep.About 1,800 words. Technical, with the arithmetic shown.
- How far is too far, and how this tool measures itWhere each ceiling sits, who set it, and what the figure beside the picture is counting when it climbs.About 2,000 words. Half argument, half specification, and five questions at the end.
- Why nothing here reshapes a face or a bodyAn absent feature explained as a decision: the line it is drawn on, and what got built instead of a warp mesh.About 1,600 words. Almost no engineering in it at all.
Two that are owed and not yet written
The first is on working a group portrait one face at a time, which is what this release requires of you and is more laborious than it ought to be. Until there is a detector in the build, the region is placed by hand and the layer stack belongs to whichever face it is currently sitting over; there are ways to do that efficiently and none of them are written down yet.
The second is on the border between retouching a person and tidying the frame around them. Taking a cable, a bin or a burnt-in date out of a background is a different operation with different failure modes — it reconstructs a region from its surroundings rather than conditioning skin — and the brush here is not built for it. That distinction deserves a page rather than a sentence, and it does not have one.
Listing both here is a commitment rather than a tease. Neither is drafted; when one is, it appears in the list above and in the row at the top of every page.