Vision model
AI that looks at a photo and reports what it sees, such as an object, printed text or a condition, as a description or as structured fields.
Updated October 2026 · with examples from our own builds
What does vision model mean in practice?
Vision models take an image and answer a question about it: which plant this is, what is printed on this card, how bright this corner of a room is, which buttons are on this screen. Many general-purpose models now accept images alongside text.
They impress in a demo and need care in production. Photos vary in light, angle and script, models misread small text and handwriting, and every image is a paid call, often counted as many tokens. The pattern that holds up is to let the model read and let code check whatever can be checked.
Before choosing a model, test it on images like the ones your users will actually take, bad ones included.
Vision model at a glance
- Reads
- Objects, printed text, scenes and screens
- Weak spots
- Handwriting, small text and unusual scripts
- Pattern that works
- The model reads, code checks
Where has vision model come up in our builds?
A photo of a spot is sorted into one of five light classes by a vision model, then blended with the window's compass direction, the placement type and the local weather. Placement scoring itself is deterministic, so the same plant in the same spot always gets the same answer.
Visiting cards are read by a vision model, then code checks Indian mobile and landline formats, email shape, PIN code to state and the GSTIN checksum. A QR code on the card is decoded by code and outranks the model.
A vision model looks at each screenshot of a customer's app and suggests which controls are sections worth opening; it can only reorder and prune what the explorer does.
Common questions about vision model
Can a vision model read handwriting?
Unreliably. Printed text is read far better than handwritten notes, so anything handwritten should be flagged for a person to check.
How do I know a vision feature is accurate enough to ship?
Build a test set of images like your users' real ones, check the answers by hand and score each model on the fields that matter. Decide beforehand what accuracy you need and what happens to the images it gets wrong.
Related terms
Planning a product and not sure where vision model fits? Ask us, or browse all 116 terms in the glossary.