DEMO / RECORDING
Overview & result
A short demonstration of text-to-screenshot selection. The user describes a visible object, and the selection moves to that region before showing a crop. The author says it was built with Jev 1.14.
From the original post
text-to-screenshot just say what you want and the crosshair goes and gets it.
built with jev 1.14 from @typesafeai
How Jev fits in the loop
- The user types a description of visible screen content.
- The demonstration locates a music player, a news panel and a named tweet in separate examples.
- A crosshair and selection box move to the requested region, followed by a cropped preview.
- The author credits Jev 1.14; the model request and implementation are not disclosed.
How to reproduce
- Watch the original 24-second video for the three selection examples.
- The post does not provide a public repository, installation guide or interactive demo.
Why this build matters
Shows a way to select screen content by meaning, reducing manual aiming and dragging when taking a screenshot.
Limitations
- Video reviewed in full; the implementation was not independently run.
- Jev use is explicitly stated by the author, but the exact API call and division of work are not public.
- TypeSafe describes Jev 1.14 image input and Box output as experimental; the demonstration does not establish general accuracy or production readiness.
- No public code or installation path is linked in the source post.