Find a UI element
Point Jev Box at a screenshot and return the normalized location of a button, badge, or menu.
Jev image preview
Ask Jev what an image contains
Upload one PNG or JPEG, describe the decision you need, and inspect a typed answer or a normalized bounding box.
Experimental image and Box features · PNG and JPEG input · Jev preview model
Response
Typed response
target_presenttrue · 0.92
target.boxtop · left · right · bottom
confidence0.88
Noul, Choice, or Box answer
What Jev Image does
Input
One PNG or JPEG image
Question
A focused visual task in plain language
Output
Noul, Choice, or Box answer
Model
jev-preview for the Box preview
Status
Experimental TypeSafe capability
Access
Use the playground with a site API key
How to use Jev Image
- 01
Upload an image
Choose a clear PNG or JPEG. The file is sent through the site server to the provider.
- 02
Describe the task
Ask one concrete question, such as whether a logo is visible or where the primary button is.
- 03
Inspect the answer
Read the typed answer, confidence, and Box edges. Use the result in your own workflow.
Useful visual decisions
Check an image condition
Use a Noul question to gate a workflow when a document, object, or visual marker is present.
Classify a scene
Ask a Choice question with categories that match your review queue or automation branch.
Why this image workflow is different
The value is in the answer shape: a visual request can become a small, inspectable decision object.
Typed visual answers
Build around Noul, Choice, and Box fields instead of parsing a paragraph.
Normalized Box coordinates
Box edges are fractions of image width and height, so the answer can be drawn at any display size.
Small questions work better
Break a broad visual task into simple checks so you can handle uncertainty explicitly.
Jev Image request facts
| Endpoint | POST /jev/image through the site BFF |
|---|---|
| Formats | image/png and image/jpeg |
| Modes | analyze and box |
| Box model | jev-preview / Jev 1.14 preview |
| Box answer | top, bottom, left, right, confidence |
| Image limit | Validated server-side before upload |
Jev Image FAQ
What is Jev Image?
Jev Image is an independent online interface for sending a PNG or JPEG to Jev with a focused visual question. It returns typed answers that can be inspected by a person or used by an application.
What is Jev Box?
Jev Box is a preview question type that returns the position of one target as top, bottom, left, and right fractions of the image. It is designed for locating one described object.
Can Box say that an object is missing?
A Box answer always contains coordinates. Ask a Noul presence question alongside it and ignore the Box when the target is not present or confidence is too low.
Is this the official TypeSafe website?
No. Jev Image is an independent third-party interface. TypeSafe AI builds Jev and publishes the provider documentation at typesafe.ai.
What images can I upload?
The first release accepts one PNG or JPEG per request. The server checks the media type and size before sending it to the provider.
Are image and Box features production-ready?
TypeSafe describes these features as experimental. Treat results as a prototype signal, test them on your own images, and keep a human or application-level fallback.
Try one visual question
Open the playground, upload a clear image, and start with the smallest useful question.