What the photo fills in

September 12, 2026

A photo taken at the shelf now fills in more of the form. The words printed on the box become a description, the closest category you already have is chosen for you, and the name it suggests is the thing itself again.

Open a line to read the whole story behind it.

• New The words read off your photo now fill in a description, kept as they were printed.

  Reading a label finds every line of text on a box, and until now most of it had nowhere to go: a name came out of it, a barcode if one was printed in digits, and the rest was something to open and read rather than to keep. Adding something new on a phone now has a description, folded away like the barcode and the part number beside it, and the text read off your photo drops into it and springs it open. The lines stay in the order they appear on the packaging, because that is the one piece of structure the picture had — the cooking time under the name, the storage note under that. Only the lines the reader was confident about go in, so a smudge it scored badly stays out of a field you are about to save, and the full list is still there to open if you want to copy something it left out. It is an ordinary editable box: keep the useful lines, delete the rest, or write something else entirely.

• New Cataloguing at the shelf now lets you pick a category, and the closest one you already have is chosen from what the photo showed.

  Adding something new on a phone could not file it anywhere. Categories were a desk job, so everything catalogued at a shelf landed in nothing and had to be sorted out later. That screen now has a category picker, listing the categories you already have with their sub-categories under them. Where a photo has been read, the kind of thing the picture appeared to show is matched against your own list and the closest one is already chosen, with a line underneath saying where that came from. It only ever picks something you already have: a guess matching nothing leaves the picker on no category rather than inventing one, because a category nobody chose to create is a catalogue quietly splitting in two. One category here, which is the question worth asking while standing at a shelf — a product can be put in as many as you like from a desk.

• New Reading a photo no longer makes you wait on one screen: the photos queue up, and you pick one when it is ready.

  Reading a photo can take a moment or two, and until now the page simply sat there for as long as it took, with nothing to do but wait — and on a slower copy it could sit there long enough to give up altogether. Now the photo is put in a queue and you land on a list of what is being read. Take the next photo straight away, put the phone in your pocket, or close the screen and come back later; when one is done it says so, and you tap it to open the new product form with its fields already filled in. Whatever you had scanned before taking the picture — the barcode, the shelf you were standing at — comes with it, so nothing is asked for twice. The list checks itself every couple of seconds and stops when there is nothing left to wait for; with scripting turned off there is a Check again button that does the same thing. It is on the phone and in an ordinary browser window both, because a photo taken at a shelf is very often looked at from a desk.

• Fixed A photo taken on a phone suggests a name for the thing itself again, rather than a stray line of text off the label.

  The name offered after reading a photo is meant to be what the thing appears to be — pandan rice, rather than whichever words happened to be longest on the packaging. Photos taken on a phone were not getting that. They came back with a fragment of label text instead, and the line under the box said it had been read off the picture rather than described, which was true but not what anybody wanted. A photo taken with a phone is far bigger than one picked from a folder, and past a certain size the step that describes the thing was refusing the picture outright and saying nothing at all. The reading still finished, so nothing looked wrong, and the name quietly fell back to the text. Photos are now sized for that step before it is asked, so a picture taken at the shelf is answered exactly as one chosen at a desk. It is quicker too, because the picture it looks at is smaller.