Updated

WeVisOne-VL

Tencent’s WeChat Vision introduced WeVisOne-VL, a vision-language model that combines object-level tasks such as detection, localization, counting and OCR with pixel-level segmentation and matting. The open model family includes 2B and 4B variants.

By No Code MBA · Facts checked

Provider
Tencent
Announcement date
Status at last check
Generally available
Model family
WeVisOne-VL
Capabilities
Text, Image

What changed

WeVisOne-VL is a vision-language model family from Tencent WeChat Vision, released in 2B and 4B sizes and built on Qwen3-VL. It handles detection, referring localization, counting, pointing and OCR, plus segmentation and alpha matting through a 9.6M-parameter pixel decoder that reuses the model's own ViT features instead of SAM. Tencent says it trained on about 100M images and reports leading results on more than 30 object- and pixel-level benchmarks, and that it can reject queries for objects that are not present.

Access and availability

Weights are on Hugging Face at tencent/WeVisOne-VL-4B and tencent/WeVisOne-VL-2B, with code on GitHub (Tencent/WeVisOne-VL) under Apache 2.0 and a hosted demo Space. There is no hosted API.

Official access documentation

What you can build

Try the 2B and 4B variants on the same image set to compare structured object localization with fine-grained segmentation and matting outputs.

Limitations

Tencent says the model targets single-turn image perception, the pixel decoder rejects video input, and fine matting boundaries are sensitive to resolution, occlusion, transparency and prompt wording. It is intended for research and not for identifying individuals or making autonomous high-impact decisions.

Sources and release history