A frame now keeps the whole sentence you said. The shutter speed and the aperture used to be taken out of it, which left the rest of the line reading oddly. They stay in, and the readings still sit on the line above. The same sentence goes into the caption when you tag your scans on the desk, so what reaches Lightroom is what you actually said.

Aperture and shutter speed on the logging screen. You can now set both without leaving the frame you are logging. The same three dials the frame editor uses sit under the sentence. What you said lands in them as it is heard, and a dial you turn by hand wins over the words, so a frame can be logged without saying anything at all. The shutter dial is in whole seconds. Bulb is no longer a position on it. Saying "bulb, four minutes" now records four minutes rather than the word, so a hand timed exposure is a number you can correct to the second. The dial runs in whole seconds to five minutes and in whole minutes to an hour. Hearing that actually works. The larger speech model has never once run. The routine that closed the microphone wiped the recording before it could be handed over, so every frame ever logged got the phone's first guess and nothing else. That is fixed. When there is signal your sentence is read by a model that knows what an f-stop is, and the words come back right. Nothing rewrites what you said. An earlier version passed the sentence through a language model to tidy it, and that model removed ordinary words. It is gone. What you said is what the log keeps. Hesitation still comes off, um and uh and one leading so or okay, but by a plain tested rule rather than a model. The end of your sentence is kept. The recogniser was being shut down the instant your thumb came up, before it had delivered the last words it had already heard. It is now given a moment to finish. Said once, recorded once. When the recogniser restarted its guess mid sentence, the two halves were glued end to end and the sentence appeared twice, the first copy cut off. The seam is now examined rather than assumed. The field while you talk. Holding the key turns the field into the recording itself, your voice moving across it with the time you have been talking. Nothing appears while you speak. The sentence arrives once, whole, when you let go, with the readings already in the dials. The words of the craft. Both the recogniser and the model are primed with photographic vocabulary, including the zone system and zones one to ten, the processes, and the print terms. Also: the say key now feels like a shutter release under the thumb, and there is no switch for the hearing any more. Where your voice goes is said plainly on your first roll and lives in the roll menu.

When the scans come back, the phone now tells you exactly what to do: send the roll file to the computer the scans are on, open bulb.film/desk there, drop both. The desk address has a copy key, and copied on the phone it pastes on your Mac. Notes no longer keep run-together label words or bare numbers left over from a misheard reading.