Name anything in a photo to find it, count it, or cut it out with clean edges

You give it a photo and a short description of what you want, in plain words. Depending on the task you pick, it draws boxes around every match, counts them, puts a point on each one, outlines the exact shape, or cuts the object out with a soft-edged mask that keeps fine detail like hair and fur. It can also read all the text in an image and show where each word sits. Ask for "the boy holding the microphone" and it picks that one person, not everyone on stage.

If the thing you describe is not in the picture, it says so instead of boxing something random. In the team's tests it turned down missing targets 96 percent of the time, where the best comparable open model managed 73 percent. It also beat SAM-based cutout tools on standard background removal benchmarks, without needing SAM at all.

It comes from Tencent's WeChat Vision team, builds on Qwen3-VL, and is released under Apache 2.0, so you can use it in commercial work. A smaller 2B version is also available.

Open the WeVisOne-VL-4B Space, which is linked from Tencent's own repository. Upload a photo, pick a task from the dropdown and type what you are after. Start with Matting: drop in a portrait with loose hair, type "person", and set the display to Green Screen to see how cleanly the edges come out. Then switch to Counting (points) on a crowded shot, like a shelf of products or a bowl of fruit, and check the number against your own count. Finally, try Grounding with something that is not in the picture and watch it answer None. To run it yourself, clone the Tencent/WeVisOne-VL repository on GitHub, install the package and call one predict function per task. You will need a graphics card.

Edges can still go wrong on low-resolution images, heavy overlap, glass and other see-through material, and results can shift with how you word the request. It works on single still images, not video.

• Remove the background from product shots or portraits without a separate cutout tool.
• Count items in a photo: stock on a shelf, people in a room, cells in a sample.
• Pick out one specific object in a busy scene just by describing it.
• Find where a word or label appears in a photo of a sign, menu or screen.
• Make training masks for your own vision project from plain text prompts.

Try the demo

View model page

Read this issue on neuralexpedition.com →