image-vision

github:wangyikun123/image-vision · first seen 2026-10-31 · source · badge

用阿里云 DashScope 视觉模型识别图片(默认 Qwen3.7-Plus,可通过 VISION_MODEL 环境变量随意更换为 qwen-vl-max 等)。当用户提供图片文件、截图、base64 图片,或要求看图/识图/分析图片内容/读图/OCR 时使用。主会话模型永远不变——图片识别是一次直接旁路调用 DashScope,结果返回后继续用用户当前模型。绝不

Skill Health: not yet scored

Listed, not yet scored. This skill has no public model test yet. It is in the scoring queue; submitted skills go first, then the most-installed.

Version history