What it does
This is an independent image-native decision model inspired by Jev’s contract, not TypeSafe Jev.
Benchmarks & research
visual-jev
Ask typed questions directly about images through a shared Qwen3-VL encoding instead of first producing captions.
Bookmark: visual-jev
Keep this in your collection.
Leave a noteWhat would you try with this?
: visual-jev
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
Start by collecting sample pictures alongside the exact questions you need answered about them. You can decide whether each question asks for a score or picks from specific choices, without having the app write sentences first.
A developer can install and run the program on your computer. Keep in mind that the system stops completely if any required model file is missing or altered, so your setup must include the exact approved files.