Zoom into the jacket and the twill sharpens: the diagonal wale runs clean, a slub sits where the yarn thickened, the topstitching holds a gauge you could almost count. None of it was photographed. GarmentZoom, a system from the University of Washington, invents that close-up from one full-view shot and a single unaligned reference, separating how good a garment looks from how it was built. In one example it lifts a five-megapixel image 12.8 times into 1.2 billion pixels of seamless zoom, and it holds detail across a continuous range from three to twenty times.

The method is reference-guided, not pure fabrication, and the quality is real. But its authors name the failure in plain terms: where the needed texture is missing from the reference, the model “must hallucinate plausible details.” A garment shot from one close-up has regions that no reference crop reached, and those are the regions the model dresses in convincing weave it was never shown. Plausible is the operative word. The zoom optimizes for what reads as quality — not for what the seam does when you pull it.

The moment a shopper decides a garment is well made now happens inside a picture the garment never sat for.

We have watched a zoom invent detail and pass as capture before. In 2023 a Reddit user showed that Samsung’s “Space Zoom” was painting crater detail onto a deliberately blurred moon, its “detail improvement engine” supplying texture the lens never resolved. Catalog teams will read GarmentZoom the same generous way, as a sharper camera. It is closer to a sharper illustrator.

Distance selling has always let the picture outrun the object. Richard Sears built the mail-order catalog on catalog illustrations, selling everything from clothes to farm machinery to rural buyers who could not feel the cloth. He underwrote the gap the way the mail-order trade always had: satisfaction guaranteed or your money back. The flattering image and the return policy arrived together — one paid for the other. GarmentZoom sharpens the image by an order of magnitude and leaves the second half of that bargain untouched.

This matters because clothing is the most returned category in online retail, and the frames that calm a shopper about make and quality are the texture close-ups. Synthesize the close-up and the quality verdict moves into a frame the cloth never faced. The parcel becomes the place the truth arrives, which is the most expensive place to put it. A product-imaging firm testing AI-generated on-model photography reached the same structural worry: hallucinations are inherent, and unfaithful images feed returns. The technology closes the detail gap on screen and reopens it at the door.

The engineering is sound. The unease is not about competence. A close-up is a promise about the hand of a fabric, the weight of it, the way a seam sits after the third wash, and this one is composed by a system that has never touched cloth. Sell that promise well enough and the garment underneath stops being the thing you are buying. It becomes the thing you hope it matches.