⬢github Jupyter Notebook · 929 ★ +121 since we first saw it · pushed 10 d ago · NOASSERTION
Tencent/WeVisOne-VL
WeVisOne-VL is a powerful VLM by the WeChat Vision Team at Tencent, unifying object-level and pixel-level perception from points and boxes to masks and alpha mattes.
WeVisOne-VL is a vision-language foundation model from Tencent's WeChat Vision team that handles many visual perception tasks in one model: detection, referring localization, counting, pointing, OCR, segmentation, and alpha matting. It uses structured next-token prediction plus a small 9.6M-parameter pixel decoder (no SAM needed), and can refuse to answer when a target isn't present.
Why now: It just released (Oct 2026) with model weights (2B & 4B), demos, a technical report, and strong benchmark numbers vs Qwen3-VL-style baselines, plus the open-sourced Mask2Alpha matting annotator.
Who it is for: Developers and researchers building vision pipelines that need unified detection, segmentation, and matting from open weights.
Stars over our 65 snapshots: 808 to 929, since 17 h ago.
Where people talked about it
- ⬢github new repos, most starred 48 min ago
API: https://socialmediatrends-api.osmike.com/v1/repos/Tencent/WeVisOne-VL