Search the text in your photos and videos
A lot of what you remember about a shot is written in it: the "For Sale" sign outside the house, the shop name, the banner at the event, the title slide of a talk. Reelary reads that text in every photo and video frame, on your machine, so typing the words finds the shot.
What it reads
- Signs and storefronts: real estate signs, shop names, street signs, menus.
- Titles and lower thirds: title cards, burned-in captions, name straps in interviews.
- Events and talks: banners, stage screens, presentation slides, whiteboards.
- Products and labels: packaging, bottles, boxes, printed tags.
- Photo folders: screenshots, scanned documents and posters sitting next to your footage.
How it works
After a folder is indexed, Reelary reads the text in the background: each photo once, and each video at the same frames it already indexes, one every two seconds. It reads the small preview frames the index already made, so it does not decode your videos again, and it uses two CPU threads so the rest of the machine stays responsive. Files that have been read are not read again unless they change.
The text model is PaddleOCR's PP-OCRv6 (Apache 2.0), about 6 MB. It downloads by itself the first time the app opens and then works offline. It reads Latin-script languages and Chinese.
Searching it
Type the words into the normal search box. A match on the text shows up next to the visual results, with the text Reelary read shown on the result, and the inspector lists everything read from the selected file, with timecodes for videos.
You can describe the shot the way you remember it. Words like "sign", "poster" or "label" are understood as where the text is, not as text to look for, so "for sale sign" finds a sign that only says FOR SALE. Pair it with what else is in the picture and the visual search does the rest.
What to expect
Large, clear text is read reliably. Very small text in a wide shot, heavy motion blur, handwriting and decorative fonts are hit or miss, and a caption that flashes up between two indexed frames of a video may not be seen. Text search matches the words that were read, so a misread word will not match. Visual search still works on those shots.
Private by design
Photos and videos are read on your machine and the text never leaves it. Reelary has no account and no cloud index; the only request this feature makes is the one-time model download. See the privacy policy for every request the app makes.
Common questions
Is on-screen text search free?
Yes. Reading text in photos and video frames is part of the free version, with no limit on how many files it reads.
Does it upload my photos or videos to read the text?
No. The text model runs on your own machine and the text it finds is stored in your local library. The only network request is the one-time download of the model itself, about 6 MB.
What text does it miss?
Large, clear text reads best. Very small text in a wide shot, heavy motion blur, handwriting and decorative fonts can be missed or misread, and text that is only on screen between two indexed frames of a video may not be seen.
Does it change my files?
No. Your photos and videos are only read. The recognised text lives in the Reelary library, next to the search index.
More in the full FAQ.
On-screen text search is free, like visual search and scene descriptions. Pro adds what was said, who is on screen and where it was shot. See Places and dialogue search.