How the photos are organised
Every photo in this album was looked at once by an AI model, and a few more things were measured directly from the
photo's colours. The results are saved with the photo, so when you filter and sort, nothing is being worked out on the
spot: the album simply looks up what's already known. That's why the filters respond instantly.
The AI: Jev
The AI used here is Jev, made by a company called TypeSafe. Jev isn't a chatbot and doesn't write
descriptions. Instead it answers specific questions with a short, fixed kind of answer, such as yes or no,
pick one from this list, or how much, and it says how sure it is of each answer.
It's used for the things that need an understanding of what's in the picture, which simple measurements can't give:
- Primary subject: what the photo is mainly of, such as water, mountains, wildlife or people.
- In the photo: whether there are people, wildlife, birds, insects, flowers, tents, vehicles or structures.
- Activities: what any people are doing, such as hiking, skiing or paddling.
- Weather & sky: clear, cloudy or foggy, and sunsets, night skies and rainbows.
- Landscape: roughly how much of the photo is sky, water, forest, snow, rock and so on.
- Mood and Look & composition: things like how dramatic the sky is, how wild the scene feels, and whether it's a close-up or a wide view.
- Quality: an overall score from 1 to 100, as a judge at a nature photography club competition might rate the photo, taking in focus and exposure, composition, light and how memorable it is. Close-ups, textures and abstract photos are judged on their own terms, not against grand views.
Each photo is asked about 50 of these questions in one go. When Jev isn't confident about an answer, the photo is
marked as unsure rather than guessed, so it only shows up when that filter is left at Any. For people,
animals and other things in the photo, Jev can also point out where they are: Show boxes outlines the most
prominent one of each it found.
Measured from the colours (no AI)
Some things don't need an AI at all: they can simply be counted. Every photo is made of tiny dots of colour (pixels),
and counting them gives exact, consistent answers. The Palette chips, the
Colour · Saturation picker, and the warmth and brightness sliders all come from counting the colours of
each photo's pixels.
A little care goes into making those counts match what you'd expect: very dark shadows and blown-out bright spots (like
the sun) are ignored, and grey counts for less than real colour, so a bright sunset over grey beach stones is called
red, not "neutral".
The date a photo was taken comes from the information the camera stores inside the file. When a photo doesn't carry
that, it shows Date: Unknown and is placed after the dated ones.
Good to know
- The AI's answers are well-informed estimates, not facts, and they can be wrong. The album's owner can correct them.
- Landscape amounts are approximate, and each is judged on its own, so they can add up to more than 100% (snow on a mountain counts as both).
- To protect privacy, when a photo is added its hidden details are removed: where it was taken (GPS location), camera serial numbers, names and contact details, and hidden preview images. Only the date taken and what's needed to show the colours correctly are kept.