What's fixed in v1.4.1
Stops the AI inventing specific names it can't actually know.
On subjects like landmarks, bridges, or buildings, a local vision model would sometimes fill a gap with a plausible-but-wrong proper name — and give a different one for each near-identical shot (four burst photos of the same bridge could come back with four different, mostly-wrong bridge names). PhotoScribe now tells the model to only name specifics that are given in your context/tags or clearly legible in the image, and to stay generic ("a bridge over a river") when it isn't sure. Better general and correct than specific and wrong.
Tip: set the Location field in the context options — that gives the model a real place name to anchor to instead of guessing.
Thanks to @Zoolander06 for the clear report (#15).
Install
- macOS — download
PhotoScribe.dmg, open it, drag PhotoScribe to Applications. Signed & notarized. - Windows — download
PhotoScribe-Setup.exeand run it. (Unsigned, so SmartScreen may warn — More info → Run anyway.)
Full history in the CHANGELOG.