Mixing real and generated photos in one listing

Asked about a single processed image, the question has a clean answer: it either represents the piece accurately or it does not. That question is answerable, and it is still the wrong one, because a single image is not how the work is consumed. The frames sit together in one listing, and whatever the buyer ends up believing about the piece is assembled out of all of them at once.

That changes what provenance is. It stops being an attribute of a file and becomes an attribute of the set. A set that contains one untouched capture can be read against that capture: every other frame is answerable to it, and a disagreement between them is visible. A set in which every frame has been through something has nothing inside it to read against, and it does not help that each individual step was reasonable when it was taken.

So the decision to make is not whether a generated frame belongs in a listing. It is which slot holds which kind of frame — and, before any of that, which slot you have decided in advance not to process at all.

Three of the processes involved are documented elsewhere on this site, and the useful thing here is not to repeat them but to note what each one takes away, because that is what decides which questions the remaining frames still have to carry.

A generated frame arrives at an output size that was settled upstream rather than chosen by you, and the page on generated image resolution covers that along with what upscaling does and does not change. A cutout arrives with a transparent background that has to be flattened onto a surface before a marketplace will accept the file, and the Etsy page covers why a cutout is not a listing photo. Flattening itself runs in one direction only, and the multi-channel page covers why the flattened file cannot become the master again.

None of that disqualifies any of these frames. It does mean each one has narrowed: the generated frame cannot be enlarged beyond what it was rendered at, and the flattened cutout no longer records what the piece was actually resting on. Write down what each frame can no longer show and a gap appears in the set. The untouched frame is what fills that gap.

Reserve one slot before you begin. In practice it is a frame you shot yourself, exported from the original file rather than from anything you prepared for upload, and left alone: no cutout, no replacement surface, no generation.

It does not have to be the best photograph in the set, and it should not be chosen for how it looks. It is chosen for what it still holds: the piece as the camera recorded it, on whatever it was actually resting on, at whatever size the camera produced. Every claim the other frames make about shape, proportion and finish can be checked against that one file.

This fails by subtraction rather than by a mistake you would notice. The cleanest frame is the one worth cutting out, so it gets cut out. The flattened export is the file you just uploaded, so it gets saved over the original. Neither step looks wrong while you are taking it, and afterwards the set has no anchor in it.

Generated frames come out of an instruction, and the instruction is not identical for every category. Knowing what it names is what lets you give a generated frame a slot it can actually fill instead of one it merely occupies.

The app works in categories — rings, bracelets, necklaces and earrings — and each is tied to a body region: hand, wrist, neck and ear. The necklace instruction names both the neck and the upper chest. Close framing is instructed for rings and bracelets; for a necklace or an earring, a close detail frame is a slot your own camera has to fill. The earring instruction is the one that carries the single-or-pair rule, so a piece listed as a pair is instructed to remain a pair. Across categories, the instruction also asks that the piece is not mirrored or duplicated and that openings, links and closures stay as they are.

Instructed is the accurate word throughout. What the app sends is an instruction, not an assurance about the frame that comes back, which is why the output is something to check rather than something to accept. The result screen carries a report path for that, with reasons that name product changed, stone count, metal colour and anatomy among others.