Compare a photograph
Use the browser classifier for an immediate, private comparison without installing anything.
Open the web classifier →Private cloud recognition
Look up.
Read the sky.
Choose a sky photo and PictureCloud compares it with eleven cloud labels directly on your device. Then use the visual atlas to make sense of the model’s leading matches.
Read texture, structure, and level
High
Middle
Low
Use PictureCloud your way
A quick comparison, a learning guide, or a pocket field tool
Start with the task that fits the moment. No account is required for the website or Android app.
Learn the visual clues
Follow a practical observation sequence, then compare the cloud guides side by side.
Read the identification guide →Take PictureCloud outside
The Android app adds camera framing, a local observation history, and an offline field guide.
Get the Android app →A simple field workflow
From sky photo to a useful comparison
The classifier provides a ranked visual comparison, while the atlas supplies the context that a score alone cannot.
Frame the cloud
Use a sharp, sky-dominant photo with visible texture. PictureCloud automatically takes a centered square crop for the model.
Analyze it privately
The photo is decoded and classified locally in your browser. It is never uploaded to a recognition server.
Compare the clues
Review the top three results, image diagnostics, all model scores, and the guide for the leading label.
Transparent by design
Know what the tool can—and cannot—tell you
PictureCloud exposes the full score distribution and publishes its measured baseline instead of presenting a prediction as certainty.
11possible output labels
2.0 MBon-device model
46.25%held-out baseline accuracy
0 uploadsfor photo recognition
Experimental visual classifier
The current model is a technically validated baseline, not a production-quality authority. It has no Clear or “unknown” class, cannot forecast weather, and must not be used for safety-critical decisions.
The classifier’s vocabulary
Ten cloud genera—and one human-made trail
The World Meteorological Organization recognizes ten cloud genera. The current model also includes contrails because they appear in its CCSN training taxonomy.
AcMiddle level
Altocumulus
Rounded patches, rolls, or cells with some shading and more apparent size than cirrocumulus.
Open the guide →AsMiddle level
Altostratus
A broad grey or bluish sheet that can reveal the Sun weakly, as though through frosted glass.
Open the guide →CbDeep vertical
Cumulonimbus
A powerful tower with a dark base and, when mature, a fibrous or anvil-shaped upper region.
Open the guide →CcHigh level
Cirrocumulus
Thin white fields of very small grains or ripples, generally without their own shading.
Open the guide →CiHigh level
Cirrus
Detached wisps, filaments, hooks, or feather-like bands with a white, silky appearance.
Open the guide →CsHigh level
Cirrostratus
A transparent, whitish veil that covers much of the sky and often produces halo phenomena.
Open the guide →CtHuman-made
Contrail
A line-shaped condensation cloud produced behind an aircraft, sometimes spreading into a broader veil.
Open the guide →CuLow base
Cumulus
Detached, dense heaps with crisp, cauliflower-like tops and a comparatively flat base.
Open the guide →NsMulti-level layer
Nimbostratus
A thick, extensive grey layer associated with widespread, persistent rain or snow.
Open the guide →ScLow level
Stratocumulus
Low fields or rolls of rounded elements, often with grey shading and breaks between cells.
Open the guide →StLow level
Stratus
A low, uniform grey layer with a fog-like appearance and sometimes light drizzle.
Open the guide →Use observation, not automation alone
A model score is the beginning of the identification
Cloud height, motion, changing shape, surrounding weather, and a second viewpoint can all matter. Compare several clues and defer to official weather services for forecasts and warnings.