Tencent's WeChat Visual Team has open-sourced the WeVisOne-VL vision-language model, unifying detection, OCR, segmentation, and image masking.
Tencent's WeChat Visual Team has released the open-source vision-language model WeVisOne-VL. It offers two parameter sizes: 2B (2.44 billion parameters) and 4B (4.83 billion parameters). Built on Qwen3-VL, the single model handles detection, referring expression generation, counting, pointing, OCR, segmentation, and alpha matting. It includes a 9.6 million parameter pixel decoder to output masks and mattes without requiring a separate SAM.
Continue looking at this topic
Direct useImage format conversion ↗Related contentWhat is Qwen-VL? How does image and video understanding differ from image generation ↗OpportunitiesCan't write code, used Codex to build a Xiaohongshu product selection and content workspace ↗Open-source projectClipboard history, OCR, and AI translation all in one: which feature should PastAi prioritize first? ↗Return to original text
A short introduction is retained here; the full text from the source website is not reposted. For specific wording and subsequent updates, please refer to the original article.
Open AIHOT original