{"install":"安装","market":"通过 dsh-market · 推荐","marketHint":"复制市场安装命令","cli":"通过命令行","copy":"复制安装命令","copyLabel":"复制安装命令","dlTitle":"下载量（近 30 天）","items":[{"cat":"vision","tag":"视觉与多模态","dl":88925,"stars":4411,"added":"2026-08-14","npm":1,"name":"modlens","owner":"liustack","short":"modlens","slug":"liustack/modlens","desc":"为纯文本模型架起视觉桥梁：粘贴图片，输出结构化 JSON 证据（OCR、版面、语义）。","cmd":"dsh plugin --profile web add @liustack/modlens","href":"/zh/p/liustack/modlens/"},{"cat":"vision","tag":"视觉与多模态","dl":30991,"stars":1166,"added":"2026-08-14","npm":1,"name":"dsh-vision-router","owner":"ysr666","short":"dsh-vision-router","slug":"ysr666/dsh-vision-router","desc":"为纯文本 Agent 提供视觉能力：内置免 Key 视觉链 + 像素级视觉工具（看图问答、定位、裁剪、像素对比、取色、OCR、矢量化、抠图、截图）；粘贴图片即可用。","cmd":"dsh plugin --profile web add dsh-vision-router","href":"/zh/p/ysr666/dsh-vision-router/"},{"cat":"vision","tag":"视觉与多模态","dl":10948,"stars":905,"added":"2026-08-13","npm":1,"name":"dsh-vision-toolkit","owner":"Anionex","short":"dsh-vision-toolkit","slug":"Anionex/dsh-vision-toolkit","desc":"让纯文本模型处理视觉任务：粘贴图片后自动切换到 Vision Toolkit 变体，支持图片问答、多图比较、长截图 OCR、截图还原前端 UI、元素定位与像素对比。默认无需 API Key——图片经作者自建的免费服务处理，每台机器每天 100 张；也可改为指向自己的服务商。","cmd":"dsh plugin --profile web add @anionex/dsh-vision-toolkit","href":"/zh/p/Anionex/dsh-vision-toolkit/"},{"cat":"vision","tag":"视觉与多模态","dl":10669,"stars":135,"added":"2026-08-16","npm":1,"name":"dsh-imagegen","owner":"dickpy","short":"dsh-imagegen","slug":"dickpy/dsh-imagegen","desc":"面向 DSH Web GUI 的 AI 生图插件：通过可配置的 OpenAI 兼容端点（gpt-image-2 / gpt-image-1 / dall-e-3）实现文生图与图生图，提供 api_url/api_key 设置卡片与侧边栏分栏生图工作台。","cmd":"dsh plugin --profile web add @dickpy/dsh-imagegen","href":"/zh/p/dickpy/dsh-imagegen/"},{"cat":"vision","tag":"视觉与多模态","dl":7699,"stars":2,"added":"2026-08-30","npm":1,"name":"dsh-image-viewer","owner":"WSL043","short":"dsh-image-viewer","slug":"WSL043/dsh-image-viewer","desc":"增强 DSH 原生图片查看体验，支持指针中心缩放、拖动、画廊、下载、键盘操作和图片区域备注。","cmd":"dsh plugin --profile web add dsh-image-viewer","href":"/zh/p/WSL043/dsh-image-viewer/"},{"cat":"vision","tag":"视觉与多模态","dl":4574,"stars":2,"added":"2026-08-26","npm":1,"name":"dsh-vision-bridge","owner":"GooDAnDReaDY","short":"dsh-vision-bridge","slug":"GooDAnDReaDY/dsh-vision-bridge","desc":"将图片交给所选的视觉模型处理（自动改写、显式工具或混合模式），使纯文本聊天模型也能处理包含图片的对话。","cmd":"dsh plugin --profile web add @goodandready/dsh-vision-bridge","href":"/zh/p/GooDAnDReaDY/dsh-vision-bridge/"},{"cat":"vision","tag":"视觉与多模态","dl":4432,"stars":4,"added":"2026-08-21","npm":1,"name":"dsh-image-pathify","owner":"dami9527","short":"dsh-image-pathify","slug":"dami9527/dsh-image-pathify","desc":"让纯文本模型也能处理聊天贴图，提供原生视觉体验，支持批量看图，走内置 OpenAI 兼容识图工具 analyze_image；具备视觉能力的模型不受影响。","cmd":"dsh plugin --profile web add dsh-image-pathify","href":"/zh/p/dami9527/dsh-image-pathify/"},{"cat":"vision","tag":"视觉与多模态","dl":4392,"stars":134,"added":"2026-08-28","npm":1,"name":"dsh-comfyui","owner":"fandc520","short":"dsh-comfyui","slug":"fandc520/dsh-comfyui","desc":"让 DeepSeek Harness 的 Agent 直接驱动本地或远程 ComfyUI：comfyui_run / comfyui_object_info / comfyui_workflow 工具生成与编辑图像、视频，附带工作流库（图工作流提取：按分量 / 主流程 / 整体）、加载区分辨率自动匹配、实时队列、SDXL 与 Wan 2.1 模板、配套 skill 与同源媒体代理。","cmd":"dsh plugin --profile web add dsh-comfyui","href":"/zh/p/fandc520/dsh-comfyui/"},{"cat":"vision","tag":"视觉与多模态","dl":3922,"stars":10,"added":"2026-09-03","npm":1,"name":"dsh-draw","owner":"PerryLink","short":"dsh-draw","slug":"PerryLink/dsh-draw","desc":"多引擎文生图（OpenAI Images 与智谱 CogView 预设）：按会话配额、引擎故障切换、凭证安全配置，结果卡片支持重新生成。","cmd":"dsh plugin --profile web add dsh-draw","href":"/zh/p/PerryLink/dsh-draw/"},{"cat":"vision","tag":"视觉与多模态","dl":3882,"stars":8,"added":"2026-09-05","npm":1,"name":"dsh-tool-vision","owner":"Scorp1o117","short":"dsh-tool-vision","slug":"Scorp1o117/dsh-tool-vision","desc":"通过 inspect_image 工具将本地图片或 HTTP(S) 图片链接发送至配置的 OpenAI 兼容视觉端点，并将文字结果带回对话。","cmd":"dsh plugin --profile web add dsh-tool-vision","href":"/zh/p/Scorp1o117/dsh-tool-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":3698,"stars":2,"added":"2026-08-26","npm":1,"name":"phone-lens#phone-lens","owner":"yxqfg","short":"phone-lens#phone-lens","slug":"yxqfg/phone-lens--packages-phone-lens","desc":"把手机相机变成 dsh 会话的实时取景与拍照输入，经局域网或 USB 直连。","cmd":"dsh plugin --profile web add phone-lens","href":"/zh/p/yxqfg/phone-lens--packages-phone-lens/"},{"cat":"vision","tag":"视觉与多模态","dl":2845,"stars":1,"added":"2026-10-14","npm":1,"name":"visionforge","owner":"LR611415","short":"visionforge","slug":"LR611415/visionforge","desc":"为 DeepSeek Harness 提供读图与生图/编辑能力（Qwen/OpenAI/Anthropic/Gemini 读图，Qwen-Image 生图与编辑）。基于 liustack/modlens（MIT）改造；请先卸载 @liustack/modlens——两者会注册同一批 vision provider 与图片拦截钩子，并存会导致读图失灵。","cmd":"dsh plugin --profile web add @lr611/visionforge","href":"/zh/p/LR611415/visionforge/"},{"cat":"vision","tag":"视觉与多模态","dl":2338,"stars":1,"added":"2026-10-16","npm":1,"name":"dsh-image-gen","owner":"copylee711","short":"dsh-image-gen","slug":"copylee711/dsh-image-gen","desc":"图像生成：侧边栏画廊页面和三个 Agent 工具（paint_image、paint_images、edit_painting）；预置 ModelScope、Google Gemini、OpenAI、硅基流动、火山方舟 Seedream、阿里云百炼、xAI、智谱等服务商，也可接入任意 OpenAI 兼容端点，每个服务商可单独设置代理。","cmd":"dsh plugin --profile web add @copylee/dsh-image-gen","href":"/zh/p/copylee711/dsh-image-gen/"},{"cat":"vision","tag":"视觉与多模态","dl":2263,"stars":36,"added":"2026-08-16","npm":1,"name":"picturereader","owner":"jing-hy","short":"picturereader","slug":"jing-hy/picturereader","desc":"给纯文本模型的\"读图\"能力：图片降分辨率+降色深+结构/色彩指纹渲染成文本网格喂回对话，模型像多模态一样自主缩放、取样、OCR 读图；纯本地零外部模型依赖，附读图方法论 skill 与可选 PaddleOCR。","cmd":"dsh plugin --profile web add picturereader","href":"/zh/p/jing-hy/picturereader/"},{"cat":"vision","tag":"视觉与多模态","dl":2205,"stars":22,"added":"2026-09-16","npm":1,"name":"dsh-agnes-studio","owner":"zmm863-commits","short":"dsh-agnes-studio","slug":"zmm863-commits/dsh-agnes-studio","desc":"一站式 AI 影视创作工作站，作为 DSH 全局浮层面板运行，不占用对话框：生图（文生图 / 图生图 / 多图合成，1K–4K、8 种宽高比）、生视频（文生视频 / 图生视频，4–12 秒、支持首帧控制）、短剧（导入 .txt/.md/.json 剧本自动拆解分镜，故事板预览与批量生成）与提示词专家工作区。零运行时依赖。","cmd":"dsh plugin --profile web add dsh-agnes-studio","href":"/zh/p/zmm863-commits/dsh-agnes-studio/"},{"cat":"vision","tag":"视觉与多模态","dl":2169,"stars":5,"added":"2026-08-16","npm":1,"name":"dsh-plugin-appshot","owner":"TaurusWood","short":"dsh-plugin-appshot","slug":"TaurusWood/dsh-plugin-appshot","desc":"DSH 版 Codex Appshots：全局快捷键精准捕获当前工作窗口，零摩擦挂载至 Composer 作为上下文向 Agent 提问。","cmd":"dsh plugin --profile web add dsh-plugin-appshot","href":"/zh/p/TaurusWood/dsh-plugin-appshot/"},{"cat":"vision","tag":"视觉与多模态","dl":2129,"stars":0,"added":"2026-09-11","npm":1,"name":"dsh-image-generation","owner":"whiteS18","short":"dsh-image-generation","slug":"whiteS18/dsh-image-generation","desc":"在设置中配置多个生图供应商，对话通过 image_generate 调用当前选中的模型；图片保存到工作区 generate/image，并在对话中内联显示。","cmd":"dsh plugin --profile web add dsh-image-generation","href":"/zh/p/whiteS18/dsh-image-generation/"},{"cat":"vision","tag":"视觉与多模态","dl":2117,"stars":8,"added":"2026-08-15","npm":1,"name":"dsh-free-vision","owner":"FuzzySoul","short":"dsh-free-vision","slug":"FuzzySoul/dsh-free-vision","desc":"免费视觉插件：纯文本模型看图 / OCR / UI / 报错分析，优先免费模型（千问 Qwen3-VL-Flash、豆包、DeepSeek-OCR），设置界面可配置。","cmd":"dsh plugin --profile web add dsh-free-vision","href":"/zh/p/FuzzySoul/dsh-free-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":2070,"stars":17,"added":"2026-08-14","npm":1,"name":"dsh-vision-proxy","owner":"Flyvhidbwo","short":"dsh-vision-proxy","slug":"Flyvhidbwo/dsh-vision-proxy","desc":"DeepSeek 大脑 + 自动识图：GUI 附加的每张图片自动经官方 deepseek-v4-flash-vision-exp 原生识图转译，再交给纯文本的 DeepSeek 作答（纯文本 V4-Pro 也能看图）；支持任意 OpenAI 兼容 VLM 与本地 Ollama 作为备选。","cmd":"dsh plugin --profile web add dsh-vision-proxy","href":"/zh/p/Flyvhidbwo/dsh-vision-proxy/"},{"cat":"vision","tag":"视觉与多模态","dl":1864,"stars":2,"added":"2026-08-19","npm":1,"name":"dsh-dseyes","owner":"Okkay712","short":"DSH-dseyes","slug":"Okkay712/DSH-dseyes","desc":"为纯文本 DeepSeek 提供原生图片上传体验：在 Web GUI 粘贴/拖入图片即以缩略图附件进入会话，发送前由 host 用免费智谱 GLM-4V-Flash 视觉 API（glm-4v-flash 降级链）读取并替换为文字描述，DeepSeek 据此回答，原图保留在聊天历史中。","cmd":"dsh plugin --profile web add dsh-dseyes","href":"/zh/p/Okkay712/DSH-dseyes/"},{"cat":"vision","tag":"视觉与多模态","dl":1700,"stars":2,"added":"2026-10-16","npm":1,"name":"dsh-vision-plugin","owner":"bug-huntter","short":"dsh-vision-plugin","slug":"bug-huntter/dsh-vision-plugin","desc":"为纯文本 DSH 模型增加可配置识图能力：在 DSH 0.1.7-rc.2 中，上传图片会完整通过适配链路，由 OpenAI 兼容视觉模型转写为文字后交给主模型；支持 OpenAI / Anthropic / Gemini / Azure 风格密钥鉴权，缺少 API Key 时会在发起请求前明确提示。","cmd":"dsh plugin --profile web add @lp181818/dsh-vision-plugin","href":"/zh/p/bug-huntter/dsh-vision-plugin/"},{"cat":"vision","tag":"视觉与多模态","dl":1683,"stars":9,"added":"2026-08-15","npm":1,"name":"dsh-deepseek-vision","owner":"siegfly","short":"dsh-deepseek-vision","slug":"siegfly/dsh-deepseek-vision","desc":"视觉语言网关：注册支持图片输入的 DeepSeek provider 路由，图片先由可配置 VL 模型（默认 Qwen-VL）描述成文字再发送。","cmd":"dsh plugin --profile web add dsh-deepseek-vision","href":"/zh/p/siegfly/dsh-deepseek-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":1491,"stars":7,"added":"2026-08-18","npm":1,"name":"dsh-screenshot","owner":"paicat1","short":"dsh-screenshot","slug":"paicat1/dsh-screenshot","desc":"DSH 轻量截图插件：轻——零依赖零二进制；摆——一键全屏、窗口排版、悬停吸附截取被遮挡窗口；自助——Agent 可自助截屏，传路径不传图，路径通用，接上 modlens（选装）一步读出结构化内容。","cmd":"dsh plugin --profile web add @paicat1/dsh-screenshot","href":"/zh/p/paicat1/dsh-screenshot/"},{"cat":"vision","tag":"视觉与多模态","dl":1364,"stars":2,"added":"2026-08-27","npm":1,"name":"dsh-tool-visual-primitives","owner":"InkshadeWoods","short":"dsh-tool-visual-primitives","slug":"InkshadeWoods/dsh-tool-visual-primitives","desc":"通过模式化提示分析对话图片（描述、UI、文档、定位、拓扑等），将含坐标原语（框、点、引用）的结构化证据以文本注入上下文，会话级缓存跨回放与压缩复用。","cmd":"dsh plugin --profile web add dsh-tool-visual-primitives","href":"/zh/p/InkshadeWoods/dsh-tool-visual-primitives/"},{"cat":"vision","tag":"视觉与多模态","dl":1357,"stars":7,"added":"2026-08-16","npm":1,"name":"dsh-vision","owner":"54xkeee","short":"dsh-vision","slug":"54xkeee/dsh-vision","desc":"纯文本 DeepSeek 的识图插件：默认走豆包 Web（零成本、免 API key，经 Windows CDP 桥接驱动已登录 Chrome），另有反重力额度（flash/pro）与 Gemini 降级通道；档位自动升级、视觉证据记忆与会话压缩恢复、内容哈希缓存、双语客户端面板。","cmd":"dsh plugin --profile web add dsh-vision-web","href":"/zh/p/54xkeee/dsh-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":1270,"stars":2,"added":"2026-08-24","npm":1,"name":"dsh-labnana","owner":"exoticknight","short":"dsh-labnana","slug":"exoticknight/dsh-labnana","desc":"为 DeepSeek Harness 接入 Labnana 图片生成：文生图 / 图生图 / 精准编辑，含积分预估、余额查询与网页设置界面。","cmd":"dsh plugin --profile web add dsh-labnana","href":"/zh/p/exoticknight/dsh-labnana/"},{"cat":"vision","tag":"视觉与多模态","dl":1267,"stars":44,"added":"2026-08-19","npm":1,"name":"dsh-design-qa","owner":"sunxin-ai","short":"dsh-design-qa","slug":"sunxin-ai/dsh-design-qa","desc":"给纯文本模型的设计稿保真判定：`deepseek_vision` 工具从任意 OpenAI 兼容视觉路由借来一只眼，让模型判断实现与设计稿是否一致——并附上支撑该判定的基准（4 组夹具、23 处注入缺陷、逐格原始输出）与其依赖的提问纪律。","cmd":"dsh plugin --profile web add dsh-design-qa","href":"/zh/p/sunxin-ai/dsh-design-qa/"},{"cat":"vision","tag":"视觉与多模态","dl":1236,"stars":2,"added":"2026-08-17","npm":1,"name":"dsh-vision#vision-route","owner":"314857493","short":"dsh-vision#vision-route","slug":"314857493/dsh-vision--packages-vision-route","desc":"注册 `deepseek-vision` 路由：Web GUI 直接粘贴图片，由免费智谱 GLM 视觉模型转译为文字后，再交给 DeepSeek 适配器作答。","cmd":"dsh plugin --profile web add dsh-vision-proxy-route","href":"/zh/p/314857493/dsh-vision--packages-vision-route/"},{"cat":"vision","tag":"视觉与多模态","dl":1181,"stars":3,"added":"2026-08-16","npm":1,"name":"dsh-vision-mix","owner":"haiziyao","short":"dsh-vision-mix","slug":"haiziyao/dsh-vision-mix","desc":"把文本、识图和生图 API 组合成一个 Mix 模型并按对话内容自动路由：纯文本走聊天模型，用户图片和 Agent 截图走识图模型，后续可继续追问同一张图；还支持图片生成与编辑及会话级调用记录。","cmd":"dsh plugin --profile web add dsh-vision-mix","href":"/zh/p/haiziyao/dsh-vision-mix/"},{"cat":"vision","tag":"视觉与多模态","dl":1160,"stars":2,"added":"2026-08-17","npm":1,"name":"dsh-vision#vision-tool","owner":"314857493","short":"dsh-vision#vision-tool","slug":"314857493/dsh-vision--packages-vision-tool","desc":"给 DeepSeek Harness 的纯文本模型补上免费识图：`vision` 工具直连智谱 GLM 视觉 API 描述/OCR 图片文件（glm-4v-flash 降级链），无需任何外部 CLI。","cmd":"dsh plugin --profile web add dsh-vision-free-eyes","href":"/zh/p/314857493/dsh-vision--packages-vision-tool/"},{"cat":"vision","tag":"视觉与多模态","dl":1155,"stars":1,"added":"2026-08-25","npm":1,"name":"aura-vision","owner":"Ck-epsilon","short":"aura-vision","slug":"Ck-epsilon/aura-vision","desc":"免费视觉识别插件：自适应切块长文档识别，支持历史收藏与 MD/Word/长图/Excel 导出。","cmd":"dsh plugin --profile web add aura-vision","href":"/zh/p/Ck-epsilon/aura-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":1146,"stars":5,"added":"2026-09-19","npm":1,"name":"dsh-omni-workstation","owner":"huashenglian","short":"dsh-omni-workstation","slug":"huashenglian/dsh-omni-workstation","desc":"面向 DSH 的全模态工作站：analyze_image 走有序多卡片 VLM 故障转移链；六个视觉工具箱工具（局部放大、色调采样、像素比对、OCR、元素检测、内联展示）共用同一套图片来源解析；generate_image 支持 OpenAI/DashScope/ComfyUI 协议；多卡片异步 generate_video 配 /build-video-tool 构建器；speak/clone_voice 语音合成覆盖七家供应商；全部由一个自动保存的设置页驱动。","cmd":"dsh plugin --profile web add dsh-omni-workstation","href":"/zh/p/huashenglian/dsh-omni-workstation/"},{"cat":"vision","tag":"视觉与多模态","dl":1127,"stars":3,"added":"2026-08-17","npm":1,"name":"dsh-vision-tools","owner":"moon09300731","short":"dsh-vision-tools","slug":"moon09300731/dsh-vision-tools","desc":"DeepSeek Harness 视觉能力全家桶：vision_understand 工具（OpenAI 兼容视觉 API，默认免费智谱 GLM-4V-Flash）+ 粘贴/拖拽/按钮三入口识图。","cmd":"dsh plugin --profile web add dsh-vision-tools","href":"/zh/p/moon09300731/dsh-vision-tools/"},{"cat":"vision","tag":"视觉与多模态","dl":1124,"stars":0,"added":"2026-08-15","npm":1,"name":"dsh-koboldcpp-hands","owner":"MicroHEROX","short":"dsh-koboldcpp-hands","slug":"MicroHEROX/dsh-koboldcpp-hands","desc":"给在线模型装上本地双手：koboldcpp_run 与 koboldcpp_vision 工具把重复的文本与视觉（OCR、图像分析、对比）劳动交给本地 KoboldCpp（llama.cpp）服务器，并按需拉起与管理服务器生命周期。","cmd":"dsh plugin --profile web add dsh-koboldcpp-hands","href":"/zh/p/MicroHEROX/dsh-koboldcpp-hands/"},{"cat":"vision","tag":"视觉与多模态","dl":1074,"stars":0,"added":"2026-09-15","npm":1,"name":"dsh-image-router","owner":"zhiwuli0228","short":"dsh-image-router","slug":"zhiwuli0228/dsh-image-router","desc":"在准入前把提示词里的图片换成视觉模型的文字分析，因此任何模型（包括纯文本模型）都能读图且会话不切换模型；另提供 describe_image 工具处理图片路径。","cmd":"dsh plugin --profile web add dsh-image-router","href":"/zh/p/zhiwuli0228/dsh-image-router/"},{"cat":"vision","tag":"视觉与多模态","dl":1070,"stars":2,"added":"2026-08-16","npm":1,"name":"dsh-youreyes","owner":"54xkeee","short":"dsh-youreyes","slug":"54xkeee/dsh-youreyes","desc":"纯文本 DeepSeek 的识图工具包：模型可主动调用的 `vision` 工具、deepseek/opencode-go 包装适配器（v4 flash/pro）、反重力额度（默认，flash/pro）/ 任意 OpenAI 兼容 VLM / Gemini / 本地 Ollama 通道、证据记忆与会话压缩恢复、内容哈希缓存、双语客户端面板。","cmd":"dsh plugin --profile web add dsh-youreyes","href":"/zh/p/54xkeee/dsh-youreyes/"},{"cat":"vision","tag":"视觉与多模态","dl":1053,"stars":1,"added":"2026-08-18","npm":1,"name":"dsh-mindseye","owner":"kanchengw","short":"dsh-mindseye","slug":"kanchengw/dsh-mindseye","desc":"为 DSH 上的纯文本模型提供视觉插件：通过分层证据记忆与缓存，实现图片理解与生成的原生交互及 GUI 自动化。","cmd":"dsh plugin --profile web add dsh-mindseye","href":"/zh/p/kanchengw/dsh-mindseye/"},{"cat":"vision","tag":"视觉与多模态","dl":1017,"stars":2,"added":"2026-08-15","npm":1,"name":"dsh-vision-recognizer","owner":"kaixinbaba","short":"dsh-vision-recognizer","slug":"kaixinbaba/dsh-vision-recognizer","desc":"视觉模型路由：把附加图片经可配置模型（15+ OpenAI 兼容与 Anthropic 供应商）转译为文字，对话仍由 DeepSeek 作答。","cmd":"dsh plugin --profile web add dsh-vision-recognizer","href":"/zh/p/kaixinbaba/dsh-vision-recognizer/"},{"cat":"vision","tag":"视觉与多模态","dl":979,"stars":5,"added":"2026-08-27","npm":1,"name":"screenshot-feedback-hook-mcp#dsh-screenshot-feedback-hook-mcp","owner":"lkh081231","short":"screenshot-feedback-hook-mcp#dsh-screenshot-feedback-hook-mcp","slug":"lkh081231/screenshot-feedback-hook-mcp--dsh-plugin","desc":"一个截图工具，外加两个默认关闭的自动截图时机。需要模型支持图片输入。","cmd":"dsh plugin --profile web add dsh-screenshot-feedback-hook-mcp","href":"/zh/p/lkh081231/screenshot-feedback-hook-mcp--dsh-plugin/"},{"cat":"vision","tag":"视觉与多模态","dl":969,"stars":0,"added":"2026-08-15","npm":1,"name":"dsh-unsloth-hands","owner":"MicroHEROX","short":"dsh-unsloth-hands","slug":"MicroHEROX/dsh-unsloth-hands","desc":"把重复的文本与视觉（OCR、图像分析、对比）劳动交给本地运行的 Unsloth Desktop（Unsloth Studio）服务器：unsloth_run 与 unsloth_vision 两个工具，纯 HTTP 客户端，不拉起也不持有任何进程。","cmd":"dsh plugin --profile web add dsh-unsloth-hands","href":"/zh/p/MicroHEROX/dsh-unsloth-hands/"},{"cat":"vision","tag":"视觉与多模态","dl":919,"stars":0,"added":"2026-08-15","npm":1,"name":"dsh-plugin-image-input","owner":"Elohia","short":"dsh-plugin-image-input","slug":"Elohia/dsh-plugin-image-input","desc":"图片转文字输入插件：粘贴/拖拽图片自动转为结构化文字描述发送，给纯文本 LLM 提供图片输入接管（OpenAI 兼容视觉 API）。","cmd":"dsh plugin --profile web add dsh-plugin-image-input","href":"/zh/p/Elohia/dsh-plugin-image-input/"},{"cat":"vision","tag":"视觉与多模态","dl":903,"stars":3,"added":"2026-08-27","npm":1,"name":"deepseek-continuity","owner":"linxuhao","short":"Deepseek-Continuity","slug":"linxuhao/Deepseek-Continuity","desc":"本地生成图像/配音/音乐/音效并听写，同一性锁定：角色、动物、道具与配音角色定妆定声一次，之后每次调用复用同一份参考；退化产物（接近纯色的图、静音的音频）被拒绝而不是当成成功返回；生成的台词可以听写回文字，好让吞掉结尾的克隆暴露出来。空闲时引擎卸载显存，生图与听写也可改指向任意 OpenAI 形状的 API 而不用本地 Vulkan 后端。","cmd":"dsh plugin --profile web add dsh-plugin-continuity","href":"/zh/p/linxuhao/Deepseek-Continuity/"},{"cat":"vision","tag":"视觉与多模态","dl":902,"stars":3,"added":"2026-08-14","npm":1,"name":"dsh-llm-vision-bridge","owner":"Einskyle","short":"dsh-llm-vision-bridge","slug":"Einskyle/dsh-llm-vision-bridge","desc":"DeepSeek 视觉桥接：注册原生 LLM provider，聊天内粘贴的图片自动由视觉模型（pi-ai/llama.cpp 上的 Qwen3-VL）转成文字描述后交给纯文本 DeepSeek 作答——图片准入、路由与会话压缩全部走 harness 原生机制，带 LRU 描述缓存与 503 重试。","cmd":"dsh plugin --profile web add dsh-llm-vision-bridge","href":"/zh/p/Einskyle/dsh-llm-vision-bridge/"},{"cat":"vision","tag":"视觉与多模态","dl":899,"stars":0,"added":"2026-09-23","npm":1,"name":"lookover#dsh-look","owner":"mengxiaoxian","short":"lookover#dsh-look","slug":"mengxiaoxian/lookover--dsh-look","desc":"macOS 场景感知探针：拉模型零轮询的 look 工具，读取最前非自身窗口（应用信息、AX 标题/选中文本、按需本地 OCR），支持召唤热键按下瞬间抓取快照注入对话。","cmd":"dsh plugin --profile web add dsh-look","href":"/zh/p/mengxiaoxian/lookover--dsh-look/"},{"cat":"vision","tag":"视觉与多模态","dl":891,"stars":8,"added":"2026-09-04","npm":1,"name":"dsh-appshots","owner":"WongYuYe","short":"dsh-appshots","slug":"WongYuYe/dsh-appshots","desc":"给 DSH Desktop 用的窗口截图，支持 macOS 和 Windows：macOS 按两边 Command、Windows 按两边 Ctrl（或点输入框旁相机）捕获当前前台窗口并附加到会话，同时把 VoiceOver 风格的辅助功能树注入为隐藏上下文。","cmd":"dsh plugin --profile web add dsh-appshots","href":"/zh/p/WongYuYe/dsh-appshots/"},{"cat":"vision","tag":"视觉与多模态","dl":888,"stars":1,"added":"2026-08-18","npm":1,"name":"dsh-vision-fallback","owner":"1HelloMan1","short":"dsh-vision-fallback","slug":"1HelloMan1/dsh-vision-fallback","desc":"将聊天图片交给固定的 OpenAI 兼容视觉模型，把事实性观察交给当前主模型，并在事件回放、上下文压缩和重启时复用会话内观察。","cmd":"dsh plugin --profile web add dsh-vision-fallback","href":"/zh/p/1HelloMan1/dsh-vision-fallback/"},{"cat":"vision","tag":"视觉与多模态","dl":884,"stars":0,"added":"2026-09-03","npm":1,"name":"dsh-vision-pro-bridge","owner":"ShaineDemo","short":"dsh-vision-pro-bridge","slug":"ShaineDemo/dsh-vision-pro-bridge","desc":"视觉桥：贴图先经 deepseek-v4-flash-vision-exp 转写为文字，再交给纯文本的 deepseek-v4-pro 回答，零第三方依赖。","cmd":"dsh plugin --profile web add dsh-vision-pro-bridge","href":"/zh/p/ShaineDemo/dsh-vision-pro-bridge/"},{"cat":"vision","tag":"视觉与多模态","dl":883,"stars":1,"added":"2026-08-18","npm":1,"name":"dsh-auto-vision","owner":"NormanFxxkingRockwell","short":"dsh-auto-vision","slug":"NormanFxxkingRockwell/dsh-auto-vision","desc":"为纯文本 DeepSeek Harness 主模型搭视觉桥：自动从已配置模型中挑选多模态模型，通过 vision 工具把图片描述以纯文本返回。","cmd":"dsh plugin --profile web add dsh-auto-vision","href":"/zh/p/NormanFxxkingRockwell/dsh-auto-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":878,"stars":0,"added":"2026-08-16","npm":1,"name":"dsh-quicksight","owner":"Isanti2016","short":"dsh-quicksight","slug":"Isanti2016/dsh-quicksight","desc":"纯文本模型双层识图：优先本地快速 OCR（RapidOCR，离线），不足时回落视觉模型（modlens）。","cmd":"dsh plugin --profile web add dsh-quicksight","href":"/zh/p/Isanti2016/dsh-quicksight/"},{"cat":"vision","tag":"视觉与多模态","dl":863,"stars":0,"added":"2026-08-15","npm":1,"name":"dsh-vision-guard","owner":"good-boy4069","short":"dsh-vision-guard","slug":"good-boy4069/dsh-vision-guard","desc":"纯文本路由的透明图片护栏：贴图不再 400 卡死会话，附带 vision_analyze 工具（OCR/PDF/docx/pptx/视频）。","cmd":"dsh plugin --profile web add dsh-vision-guard","href":"/zh/p/good-boy4069/dsh-vision-guard/"},{"cat":"vision","tag":"视觉与多模态","dl":856,"stars":3,"added":"2026-08-15","npm":1,"name":"dsh-plugin-grok2api-media-tool","owner":"lsjspl","short":"dsh-plugin-grok2api-media-tool","slug":"lsjspl/dsh-plugin-grok2api-media-tool","desc":"让 dsh 通过 grok2api 的 API 获得生成图片与视频能力。","cmd":"dsh plugin --profile web add dsh-plugin-grok2api-media-tool","href":"/zh/p/lsjspl/dsh-plugin-grok2api-media-tool/"},{"cat":"vision","tag":"视觉与多模态","dl":849,"stars":1,"added":"2026-08-15","npm":1,"name":"dsh-vision-subagent","owner":"ruby1304","short":"dsh-vision-subagent","slug":"ruby1304/dsh-vision-subagent","desc":"为任意 DSH 路由装上眼睛：Web 输入框贴图按意图自动聚焦分析，视觉子代理代读工作区图片，原图可随时物化编辑。","cmd":"dsh plugin --profile web add dsh-vision-subagent","href":"/zh/p/ruby1304/dsh-vision-subagent/"},{"cat":"vision","tag":"视觉与多模态","dl":844,"stars":0,"added":"2026-08-16","npm":1,"name":"dsh-subagent-vision","owner":"niuniuaba","short":"dsh-subagent-vision","slug":"niuniuaba/dsh-subagent-vision","desc":"让纯文本 DeepSeek 代理在同一会话内读图：委派给视觉子代理，发送时自动把图片转为文件路径。","cmd":"dsh plugin --profile web add dsh-subagent-vision","href":"/zh/p/niuniuaba/dsh-subagent-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":826,"stars":9,"added":"2026-08-16","npm":1,"name":"dsh-chat-imagine","owner":"corrinehu","short":"dsh-chat-imagine","slug":"corrinehu/dsh-chat-imagine","desc":"在 DSH 聊天窗口自动调用生图工具（API 渠道，或本机 CLI：已支持 mmx / codex / agy）并展示图片，也支持利用对应 CLI 识别图片。","cmd":"dsh plugin --profile web add dsh-chat-imagine","href":"/zh/p/corrinehu/dsh-chat-imagine/"},{"cat":"vision","tag":"视觉与多模态","dl":819,"stars":4,"added":"2026-08-15","npm":1,"name":"deepseek-vision#plugins/dsh-plugin-deepseek-vision","owner":"GOU-GEE","short":"deepseek-vision#plugins/dsh-plugin-deepseek-vision","slug":"GOU-GEE/deepseek-vision--plugins-dsh-plugin-deepseek-vision","desc":"面向纯文本 DeepSeek 的视觉 MCP + DSH 插件：analyze_image / analyze_clipboard / compare_images / vision_status 四个工具、可视化配置页、默认免费 GLM-4.6V-Flash、结果缓存与限流容错，Key 不入日志。","cmd":"dsh plugin --profile web add dsh-plugin-deepseek-vision","href":"/zh/p/GOU-GEE/deepseek-vision--plugins-dsh-plugin-deepseek-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":796,"stars":0,"added":"2026-09-15","npm":1,"name":"dsh-image-guard","owner":"mafeis","short":"dsh-image-guard","slug":"mafeis/dsh-image-guard","desc":"发送前把请求中的历史图片裁剪至保留张数，并从上游 400 错误解析单次图片上限、按更少张数降级重试，保证含图会话不被图片数量限制打断。","cmd":"dsh plugin --profile web add dsh-image-guard","href":"/zh/p/mafeis/dsh-image-guard/"},{"cat":"vision","tag":"视觉与多模态","dl":793,"stars":2,"added":"2026-08-25","npm":1,"name":"dsh-capture","owner":"Max-Null","short":"dsh-capture","slug":"Max-Null/dsh-capture","desc":"DSH 双引擎截图：思灵桌面壳内，全局快捷键/托盘打开全屏框选浮层（多显示器、逐屏像素级抓帧），选区定格后工具条就地画红框，一次确认送进输入框；纯 DSH（浏览器）下由输入框相机按钮经 getDisplayMedia 捕获屏幕，复用同一套单阶段框选+标注遮罩。均走官方附件通道。","cmd":"dsh plugin --profile web add @max-null/dsh-capture","href":"/zh/p/Max-Null/dsh-capture/"},{"cat":"vision","tag":"视觉与多模态","dl":764,"stars":4,"added":"2026-08-18","npm":1,"name":"visual-review","owner":"wang-bool","short":"visual-review","slug":"wang-bool/visual-review","desc":"在 DSH Web 聊天界面内联渲染粘贴/上传的图片，并让纯文本模型“看见”图片：visual_review 工具优先调用任意 OpenAI 兼容多模态 API，未配置时回退本机 Qwen3-VL。","cmd":"dsh plugin --profile web add visual-review","href":"/zh/p/wang-bool/visual-review/"},{"cat":"vision","tag":"视觉与多模态","dl":741,"stars":1,"added":"2026-08-15","npm":1,"name":"mimo-vision","owner":"wulusai2333","short":"mimo-vision","slug":"wulusai2333/mimo-vision","desc":"`describe_image` 视觉桥：经 opencode Zen API（凭据 `OPENCODE_GO_API_KEY`，免费线路优先、付费兜底）把图片发给 mimo-v2.5、返回文字描述给纯文本模型，原生格式直发，SVG/TIFF/HEIC 等格式经 ImageMagick 自动转码。","cmd":"dsh plugin --profile web add mimo-vision","href":"/zh/p/wulusai2333/mimo-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":740,"stars":0,"added":"2026-08-19","npm":1,"name":"dsh-autovision","owner":"Junkrat9527","short":"dsh-autovision","slug":"Junkrat9527/dsh-autovision","desc":"纯文本 dsh 模型的识图插件：粘贴图片自动由你配置的多模态模型转成文字——透明孪生路由、模型可自主调用识图工具、无内置密钥与中转。","cmd":"dsh plugin --profile web add @iroam2375/dsh-autovision","href":"/zh/p/Junkrat9527/dsh-autovision/"},{"cat":"vision","tag":"视觉与多模态","dl":700,"stars":44,"added":"2026-10-17","npm":1,"name":"dsh-imagegen-video","owner":"q253595422","short":"dsh-imagegen-video","slug":"q253595422/dsh-imagegen-video","desc":"面向 DSH Web GUI 的 AI 生图与生视频插件：文生图/图生图按渠道配置模型目录，并按模型族适配请求协议（gpt-image、Grok Imagine、Nano Banana、Seedream、GLM-Image、Qwen-Image、MiniMax image-01、DALL-E 3），提供侧边栏生图工作台、带节点工作流的无限画布、文件预览与画布技能，以及提示词模板库；生视频侧含文生视频、图生视频、首尾帧与多参考图模式，走异步提交/轮询/下载链路，另有宿主侧视频工具、画布视频节点，并可用本机 ffmpeg 把两个以上视频节点拼接成一个 MP4。","cmd":"dsh plugin --profile web add @q253595422/dsh-imagegen-video","href":"/zh/p/q253595422/dsh-imagegen-video/"},{"cat":"vision","tag":"视觉与多模态","dl":599,"stars":6,"added":"2026-10-16","npm":1,"name":"dsh-comfyui-image","owner":"wbb316","short":"dsh-comfyui-image","slug":"wbb316/dsh-comfyui-image","desc":"用本地 ComfyUI 出图：一个 generate_image 工具覆盖文生图、图生图、局部重绘与去背景。","cmd":"dsh plugin --profile web add dsh-comfyui-image","href":"/zh/p/wbb316/dsh-comfyui-image/"},{"cat":"vision","tag":"视觉与多模态","dl":593,"stars":0,"added":"2026-10-15","npm":1,"name":"dsh-ocr-free","owner":"vitas","short":"dsh-ocr-free","slug":"vitas/dsh-ocr-free","desc":"在 macOS 上使用 Apple Vision 的本地 OCR：从截图、照片和扫描件中读取文字，无需安装、无需联网、无需 API 密钥。","cmd":"dsh plugin --profile web add dsh-ocr-free","href":"/zh/p/vitas/dsh-ocr-free/"},{"cat":"vision","tag":"视觉与多模态","dl":582,"stars":1,"added":"2026-10-17","npm":1,"name":"dsh-image2-plugin","owner":"544244857","short":"dsh-image2-plugin","slug":"544244857/dsh-image2-plugin","desc":"注册 `image2_generate` 工具：调用你自己的 OpenAI 兼容图片中转站生成或编辑图片（文生图走 POST /images/generations，图生图走 POST /images/edits 并接受本地 PNG/JPEG/WebP 参考图，兼容 b64_json 与图片 URL 返回），图片保存在会话工作目录内，并借助其 `@wisdoverse/dsh-inline-media-viewer` 依赖在对话里内联渲染。","cmd":"dsh plugin --profile web add dsh-image2-tool","href":"/zh/p/544244857/dsh-image2-plugin/"},{"cat":"vision","tag":"视觉与多模态","dl":484,"stars":0,"added":"2026-10-13","npm":1,"name":"dsh-image-conatiner","owner":"aa2246740","short":"dsh-image-conatiner","slug":"aa2246740/dsh-image-conatiner","desc":"以对话内画廊和灯箱展示生成的图片。","cmd":"dsh plugin --profile web add dsh-image-conatiner","href":"/zh/p/aa2246740/dsh-image-conatiner/"},{"cat":"vision","tag":"视觉与多模态","dl":397,"stars":0,"added":"2026-10-15","npm":1,"name":"dsh-image-tiler","owner":"Ssrs-Pro-Max","short":"dsh-image-tiler","slug":"Ssrs-Pro-Max/dsh-image-tiler","desc":"让超大图片以可读分辨率进入模型视野，而不是被压成一张缩图：准入前按附件存储允许的上限把大图切成多张切片（8000x8000 的截图会变成 20 张 2000x1600），小字得以保留。以替换本地附件存储的方式实现，不新增任何配置项。","cmd":"dsh plugin --profile web add dsh-image-tiler","href":"/zh/p/Ssrs-Pro-Max/dsh-image-tiler/"},{"cat":"vision","tag":"视觉与多模态","dl":391,"stars":1,"added":"2026-10-08","npm":1,"name":"computer-use-vision","owner":"xuanyuanluoxue","short":"computer-use-vision","slug":"xuanyuanluoxue/computer-use-vision","desc":"Windows 桌面识图+模拟操作能力：截图→vision模型→模拟鼠标键盘，内置自进化应用技巧库。支持插件工具模式和纯skill模式双形态。","cmd":"dsh plugin --profile web add dsh-computer-use-vision","href":"/zh/p/xuanyuanluoxue/computer-use-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":372,"stars":6,"added":"2026-09-15","npm":1,"name":"dsh-ocr-local","owner":"grelvan","short":"dsh-ocr-local","slug":"grelvan/dsh-ocr-local","desc":"纯文本路由的本地 OCR 兜底：会话模型明确声明不接受图片时，把附件的图片存入本地缓存并注入路径，模型调 ocr_image 用 PP-OCRv5 + ONNX Runtime 在纯 CPU 上离线识别，无需 API key，图片不出本机；模型能看图时静默。","cmd":"dsh plugin --profile web add dsh-ocr-local","href":"/zh/p/grelvan/dsh-ocr-local/"},{"cat":"vision","tag":"视觉与多模态","dl":248,"stars":3,"added":"2026-10-17","npm":1,"name":"dsh-read-animation","owner":"NeoHotaru","short":"dsh-read-animation","slug":"NeoHotaru/dsh-read-animation","desc":"把 gif 和动图 webp 当作动画来读：贴进聊天框的照常播放，模型收到的是横跨整段采样、带帧号与时长的胶片条；工作区里的文件由 read_animation 按需采样。","cmd":"dsh plugin --profile web add dsh-read-animation","href":"/zh/p/NeoHotaru/dsh-read-animation/"},{"cat":"vision","tag":"视觉与多模态","dl":78,"stars":2,"added":"2026-10-22","npm":1,"name":"imgpost","owner":"hige6","short":"imgpost","slug":"hige6/imgpost","desc":"让 DSH 助手把本地、网页和 AI 生成的图片直接发到聊天里给用户看，并为文本模型提供识图能力；Windows 下经 PowerShell 运行。","cmd":"dsh plugin --profile web add imgpost","href":"/zh/p/hige6/imgpost/"},{"cat":"vision","tag":"视觉与多模态","dl":70,"stars":0,"added":"2026-10-22","npm":1,"name":"dsh-media-lens","owner":"drscrewdriver","short":"dsh-media-lens","slug":"drscrewdriver/dsh-media-lens","desc":"liustack/modlens 的全媒体（视频/音频/图像）延伸：把本地媒体转成结构化证据（时间线/转写/版面区域）注入对话，让纯文本模型能引用具体帧、秒、说话人——Evidence, not imagination。每个媒体一条 failover 链，逐跳记录在 meta.attempts（降级绝不静默）：宿主已注册全模态模型优先，其次自配 OpenAI-compatible 端点（逗号分隔多 key 轮换），再落到本地 ffmpeg 抽帧 / 本地 ASR CLI，末位仅元数据降级。吸收 modlens 实战经验——仅 auth/quota 类失败触发冷却且软化垫队尾（不剔除）、错误文本先脱密后截断、magic-byte 校验绝不信任扩展名。可选 vision-bridge 把透镜包装成模型选择器可见的 media-lens 包装 provider，让纯文本主力获得图/音/视频视野。多模态模型的注册/认证与真正的音视频直传依赖两个姊妹插件配合（均声明为 optional peer，未装则能力自动降级、不报错）：dsh-thinking-levels 承接模型配置录入/登记面，dsh-llm-openai-completions 做 OpenAI-compatible 传输接管、把 file 块序列化成 video_url / input_audio 真正上送。隐私：媒体缺省会发往远端 provider，开启 strictLocalOnly 则不出本机。不随包分发的安装前提：PATH 上的 ffmpeg/ffprobe（抽帧/重编码/时长探测，缺失时降为仅元数据），以及音频兜底需用户自备本地 ASR CLI。支持宿主 0.1.7 / 0.2.0 / 0.2.1 共 6 个 rc（engines.dsh 精确枚举）。npm 包名：dsh-media-lens。","cmd":"dsh plugin --profile web add dsh-media-lens","href":"/zh/p/drscrewdriver/dsh-media-lens/"},{"cat":"vision","tag":"视觉与多模态","dl":0,"stars":2,"added":"2026-10-22","npm":1,"name":"dsh-image-lite","owner":"yupaoa","short":"dsh-image-lite","slug":"yupaoa/dsh-image-lite","desc":"DSH 生图插件，双模式：对话中由模型调用 image_generate_lite 工具出图，或在右侧栏「生图工作台」手动生图；最多 5 张参考图、支持 OpenAI images 接口格式，密钥存于 DSH 凭据服务。","cmd":"dsh plugin --profile web add dsh-image-lite","href":"/zh/p/yupaoa/dsh-image-lite/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":26,"added":"2026-08-17","npm":0,"name":"dsh-web-ui#dsh-tool-describe-image","owner":"DamonKoy","short":"dsh-web-ui#dsh-tool-describe-image","slug":"DamonKoy/dsh-web-ui--packages-dsh-tool-describe-image","desc":"describe_image 模型工具：让纯文本模型通过视觉语言模型获得图像理解能力。","cmd":"dsh plugin --profile web add github:DamonKoy/dsh-web-ui#path:/packages/dsh-tool-describe-image","href":"/zh/p/DamonKoy/dsh-web-ui--packages-dsh-tool-describe-image/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":21,"added":"2026-08-15","npm":0,"name":"dsh-visual-plugin","owner":"jyh20030112","short":"dsh-visual-plugin","slug":"jyh20030112/dsh-visual-plugin","desc":"给纯文本模型装上眼睛：把用户图片转发给任意 OpenAI 兼容的视觉模型生成描述，并在 Web UI 右侧面板展示结果。","cmd":"dsh plugin --profile web add github:jyh20030112/dsh-visual-plugin","href":"/zh/p/jyh20030112/dsh-visual-plugin/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":21,"added":"2026-08-18","npm":0,"name":"dsh-media-skills","owner":"MJorgin","short":"dsh-media-skills","slug":"MJorgin/dsh-media-skills","desc":"面向纯文本模型的免费视觉桥与生图：粘贴读图、GLM-4V-Flash 与 Gemini 引擎故障转移、modlens 同款结构化证据输出，并自动播种免费视觉模型路由。","cmd":"dsh plugin --profile web add github:MJorgin/dsh-media-skills","href":"/zh/p/MJorgin/dsh-media-skills/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":13,"added":"2026-08-18","npm":0,"name":"dsh-vision-opencode","owner":"poiuyjie","short":"dsh-vision-opencode","slug":"poiuyjie/dsh-vision-opencode","desc":"给纯文本主模型加可配置识图模型：vision_read_image 工具、输入框识图模型选择器，以及纯文本路由的图片自动转文字。","cmd":"dsh plugin --profile web add github:poiuyjie/dsh-vision-opencode","href":"/zh/p/poiuyjie/dsh-vision-opencode/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":12,"added":"2026-08-15","npm":0,"name":"gemini-eyes","owner":"ConsoleSun","short":"Gemini-Eyes","slug":"ConsoleSun/Gemini-Eyes","desc":"接入 gemini.google.com 的 MCP 桥：图片/视频视觉识别、Imagen 生图、Veo 生视频与历史对话管理，复用浏览器登录态，无需 API Key。","cmd":"dsh plugin --profile web add github:ConsoleSun/Gemini-Eyes","href":"/zh/p/ConsoleSun/Gemini-Eyes/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":9,"added":"2026-10-18","npm":0,"name":"dsh-qwen-studio","owner":"kingcharleslzy-ai","short":"dsh-qwen-studio","slug":"kingcharleslzy-ai/dsh-qwen-studio","desc":"DSH 的 Qwen Image 2.1 创作工作区：本机 MLX 或自己的 NVIDIA 服务器，支持原生参考编辑、LoRA、结构控制、蒙版、整套预设、图库与对话实时进度，提供校验更新和 Agent 安装指南。","cmd":"dsh plugin --profile web add \"https://github.com/kingcharleslzy-ai/dsh-qwen-studio/releases/latest/download/dsh-qwen-studio.tgz\"","href":"/zh/p/kingcharleslzy-ai/dsh-qwen-studio/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":9,"added":"2026-08-14","npm":0,"name":"dsh-vision","owner":"linenxi-ctrl","short":"dsh-vision","slug":"linenxi-ctrl/dsh-vision","desc":"外挂识图插件：鲸鱼按钮配置面板、图片识图自动回传、模型自主截图识图工具。","cmd":"dsh plugin --profile web add github:linenxi-ctrl/dsh-vision","href":"/zh/p/linenxi-ctrl/dsh-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":8,"added":"2026-08-28","npm":0,"name":"dsh-highres-vision","owner":"azwosile","short":"dsh-highres-vision","slug":"azwosile/dsh-highres-vision","desc":"为 DeepSeek Harness 原生视觉模型 deepseek-v4-flash-vision-exp 提供高清识图：将图片准入上限提升至 32 MiB / 8192 px / 600 张，并新增 highres_read 工具，将大图切为整图与 800x800 分块后经宿主 read_image 注入模型。","cmd":"dsh plugin --profile web add github:azwosile/dsh-highres-vision","href":"/zh/p/azwosile/dsh-highres-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":7,"added":"2026-08-26","npm":0,"name":"dsh-hos-scrcpy","owner":"ns-zzj","short":"dsh-hos-scrcpy","slug":"ns-zzj/dsh-hos-scrcpy","desc":"在 DSH 网页中用 AI 操控鸿蒙手机：实时 H.264 投屏、触控与系统按键、hilog 日志，并让模型读屏、定位 UI 控件，再执行点击/长按/按键/输入。","cmd":"dsh plugin --profile web add github:ns-zzj/dsh-hos-scrcpy","href":"/zh/p/ns-zzj/dsh-hos-scrcpy/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":7,"added":"2026-09-19","npm":0,"name":"deepseek-harness-plugins#vision-bridge","owner":"Yinxe","short":"deepseek-harness-plugins#vision-bridge","slug":"Yinxe/deepseek-harness-plugins--plugins-vision-bridge","desc":"让纯文本模型也能看图：消息中出现图片占位符时模型调用 vision_describe 工具，插件把本轮图片引用与提问一起转交给多模态视觉模型识别，主模型失败时自动用备用模型重试。配置在设置页完成并通过官方 settings API 持久化到 settings.yaml，热重载不丢。","cmd":"dsh plugin --profile web add github:Yinxe/deepseek-harness-plugins#path:/plugins/vision-bridge","href":"/zh/p/Yinxe/deepseek-harness-plugins--plugins-vision-bridge/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":6,"added":"2026-08-27","npm":0,"name":"dsh-pdf-reader","owner":"AngelosZou","short":"dsh-pdf-reader","slug":"AngelosZou/dsh-pdf-reader","desc":"面向视觉模型的内容感知 PDF 读取插件：逐页分析图（矢量与栅格）、表格、公式风险与双栏排版，基于内容识别进行混合提取，图表/公式页渲染成高 DPI 区域裁切。提供低清预览以理解页面概况并支持指定选区获取高清渲染。功能包装为多个工具供Agent使用。","cmd":"dsh plugin --profile web add github:AngelosZou/dsh-pdf-reader","href":"/zh/p/AngelosZou/dsh-pdf-reader/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":5,"added":"2026-10-18","npm":0,"name":"dsh-image-preview","owner":"algerkong","short":"dsh-image-preview","slug":"algerkong/dsh-image-preview","desc":"统一接管 DSH 会话里的图片预览：read_image 缩略图，以及用户、助手、Markdown 和 present 图片，支持缩放、旋转、拖动、下载、侧栏和原生文件操作的 Lightbox。","cmd":"dsh plugin --profile web add github:algerkong/dsh-image-preview","href":"/zh/p/algerkong/dsh-image-preview/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":5,"added":"2026-08-17","npm":0,"name":"dsh-guide-dog","owner":"AtropinolTT","short":"dsh-guide-dog","slug":"AtropinolTT/dsh-guide-dog","desc":"基于 MiniMax 的多模态插件：实时语音通话模式（流式对话、悬浮胶囊 UI）、语音模式与麦克风语音输入，并提供图像/视频/音乐/语音生成与视觉检查工具。","cmd":"dsh plugin --profile web add github:AtropinolTT/dsh-guide-dog","href":"/zh/p/AtropinolTT/dsh-guide-dog/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":5,"added":"2026-08-15","npm":0,"name":"dsh-image-gen","owner":"LeemanCheung","short":"dsh-image-gen","slug":"LeemanCheung/dsh-image-gen","desc":"GPT Image 2 `image_gen`：默认复用 Codex 订阅 OAuth，也可显式使用 API Key；显影卡片、最多 3 张 API 实时局部图、持久附件回放/灯箱/下载、纯文本模型输出和受限的凭据安全请求。","cmd":"dsh plugin --profile web add github:LeemanCheung/dsh-image-gen","href":"/zh/p/LeemanCheung/dsh-image-gen/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":4,"added":"2026-08-27","npm":0,"name":"phone-eye","owner":"boheastill","short":"phone-eye","slug":"boheastill/phone-eye","desc":"让 AI 助手看见并操作真实安卓手机：phone_look（视觉+UI 树融合判读）、点击/滑动/输入、截图——经 adb，适用于任何 MCP 客户端。","cmd":"dsh plugin --profile web add github:boheastill/phone-eye","href":"/zh/p/boheastill/phone-eye/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":4,"added":"2026-08-17","npm":0,"name":"dsh-vision-bridge","owner":"GXX182","short":"dsh-vision-bridge","slug":"GXX182/dsh-vision-bridge","desc":"将会话图片转交可配置视觉服务，并向符合条件的 DeepSeek Harness 模型路由返回纯文本分析。","cmd":"dsh plugin --profile web add github:GXX182/dsh-vision-bridge","href":"/zh/p/GXX182/dsh-vision-bridge/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":4,"added":"2026-08-15","npm":0,"name":"free-vision-skill","owner":"niyongsheng","short":"free-vision-skill","slug":"niyongsheng/free-vision-skill","desc":"基于 macOS Vision Framework 的全本地识图插件：`ocr_image`（文字提取，表格结构+坐标）与 `view_image`（场景/人脸/二维码）；web 输入框可直接粘贴多张图片，图片永不离开你的 Mac。","cmd":"dsh plugin --profile web add github:niyongsheng/free-vision-skill","href":"/zh/p/niyongsheng/free-vision-skill/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":4,"added":"2026-08-15","npm":0,"name":"dsh-draw-router","owner":"xiaozhe7772222","short":"dsh-draw-router","slug":"xiaozhe7772222/dsh-draw-router","desc":"DeepSeek Harness 统一绘图路由器：自动识别任意 OpenAI 兼容接口的绘图模型，提供 draw_image 和 draw_list_sources 工具，支持商汤、阶跃、Agnes、千问、Flux、SD、Imagen 等。","cmd":"dsh plugin --profile web add github:xiaozhe7772222/dsh-draw-router","href":"/zh/p/xiaozhe7772222/dsh-draw-router/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":4,"added":"2026-08-23","npm":0,"name":"dsh-vision-hub#tool-vision","owner":"xing666173","short":"dsh-vision-hub#tool-vision","slug":"xing666173/dsh-vision-hub--tool-vision","desc":"增强版视觉工具箱:14 个像素级视觉工具(看图问答/定位/检测/裁剪/像素比对/OCR/长截图OCR/矢量化/取色/抠图/截图/展示/落盘/网页截图),单一 OpenAI 兼容端点驱动,桥接标记极简、规则下沉系统提示,带内容安全识别与限流自动重试。","cmd":"dsh plugin --profile web add github:xing666173/dsh-vision-hub#path:/tool-vision","href":"/zh/p/xing666173/dsh-vision-hub--tool-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":3,"added":"2026-10-15","npm":0,"name":"dsh-comfyui-image","owner":"QXingYShu","short":"dsh-comfyui-image","slug":"QXingYShu/dsh-comfyui-image","desc":"通过本地 ComfyUI 的 HTTP 接口把工作流排到用户自己的 GPU 上，由 Agent 直接生图：Z-Image-Turbo 适合快速草稿，Qwen-Image-2.1 适合成图与图中文字排版。 注册 comfyui_generate、comfyui_status 两个工具和一个涵盖模型选择与提示词 写法的 skill，无运行时依赖；没有 ComfyUI 在跑时会自动 headless 启动一个。","cmd":"dsh plugin --profile web add github:QXingYShu/dsh-comfyui-image","href":"/zh/p/QXingYShu/dsh-comfyui-image/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":3,"added":"2026-08-15","npm":0,"name":"dsh-plugin-multimodal","owner":"shinjiyu","short":"dsh-plugin-multimodal","slug":"shinjiyu/dsh-plugin-multimodal","desc":"在纯文本 DeepSeek 线路上开放贴图准入，用视觉 sidecar 把附件转成文字，原生视觉模型不改写。","cmd":"dsh plugin --profile web add github:shinjiyu/dsh-plugin-multimodal","href":"/zh/p/shinjiyu/dsh-plugin-multimodal/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":3,"added":"2026-08-15","npm":0,"name":"dsh-plugins#dsh-vision-tool","owner":"TZHR-invest","short":"dsh-plugins#dsh-vision-tool","slug":"TZHR-invest/dsh-plugins--packages-dsh-vision","desc":"Agent 可调用的视觉工具：通过你配置的任意 OpenAI 兼容视觉端点描述本地图片，支持多模型交叉核对，不内置任何密钥。","cmd":"dsh plugin --profile web add github:TZHR-invest/dsh-plugins#path:/packages/dsh-vision","href":"/zh/p/TZHR-invest/dsh-plugins--packages-dsh-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":3,"added":"2026-08-23","npm":0,"name":"dsh-image-vision","owner":"xiaoyuink","short":"dsh-image-vision","slug":"xiaoyuink/dsh-image-vision","desc":"让任意 DSH 模型都能识图：提供 vision/OCR/定位/裁剪四件套工具，内置病理、细胞、解剖、临床与科研统计图等专业预设。","cmd":"dsh plugin --profile web add github:xiaoyuink/dsh-image-vision","href":"/zh/p/xiaoyuink/dsh-image-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":2,"added":"2026-08-19","npm":0,"name":"dseyesopen","owner":"balue8246-maker","short":"dseyesopen","slug":"balue8246-maker/dseyesopen","desc":"视觉桥：纯文本 DeepSeek 也能收图（单独发、图文混发均可），快速小识图模型后台描述，对话框里图还在、模型只见文字；纯插件，卸载即复原。","cmd":"dsh plugin --profile web add github:balue8246-maker/dseyesopen","href":"/zh/p/balue8246-maker/dseyesopen/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":2,"added":"2026-09-18","npm":0,"name":"dsh-gpt-image","owner":"CAI-MH","short":"dsh-gpt-image","slug":"CAI-MH/dsh-gpt-image","desc":"控制已登录的网页版 ChatGPT 生成图片并下载到本地。","cmd":"dsh plugin --profile web add github:CAI-MH/dsh-gpt-image","href":"/zh/p/CAI-MH/dsh-gpt-image/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":2,"added":"2026-08-16","npm":0,"name":"dsh-deepseek-vision","owner":"Cheng-cheng9669","short":"dsh-deepseek-vision","slug":"Cheng-cheng9669/dsh-deepseek-vision","desc":"复用 DeepSeek 网页端识图模式给纯文本模型补上看图能力：deepseek_vision 工具调用本地 deepseek-vision-cli 浏览器自动化（手动登录助手、自动开启深度思考、用后自动关窗），把图片描述以文字返回给模型。","cmd":"dsh plugin --profile web add github:Cheng-cheng9669/dsh-deepseek-vision","href":"/zh/p/Cheng-cheng9669/dsh-deepseek-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":2,"added":"2026-08-17","npm":0,"name":"dsh-open-eyes","owner":"hyper-dsh-plugins","short":"dsh-open-eyes","slug":"hyper-dsh-plugins/dsh-open-eyes","desc":"为纯文本 DeepSeek 路由提供视觉桥接：通过可配置的 OpenAI Responses、Chat Completions 或 Anthropic Messages 端点分析附件与本地图片，同时保持支持图片的路由走原生路径。","cmd":"dsh plugin --profile web add github:hyper-dsh-plugins/dsh-open-eyes","href":"/zh/p/hyper-dsh-plugins/dsh-open-eyes/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":2,"added":"2026-08-28","npm":0,"name":"dsh-visibridge","owner":"lhbsaa","short":"dsh-visibridge","slug":"lhbsaa/dsh-visibridge","desc":"结构化视觉证据（OCR/版面/语义）+ USB 摄像头拍摄工具，支持\"拍-看-改\"调试闭环；后端支持 Ollama、DeepSeek、小米。","cmd":"dsh plugin --profile web add github:lhbsaa/dsh-visibridge","href":"/zh/p/lhbsaa/dsh-visibridge/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":2,"added":"2026-08-24","npm":0,"name":"taxue-dsh-artisan","owner":"taxueseek","short":"taxue-dsh-artisan","slug":"taxueseek/taxue-dsh-artisan","desc":"DSH 一体化视觉创作工具链：提示词反推与审计优化，多供应商生图，支持异步后台出图。","cmd":"dsh plugin --profile web add github:taxueseek/taxue-dsh-artisan","href":"/zh/p/taxueseek/taxue-dsh-artisan/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":2,"added":"2026-08-16","npm":0,"name":"dsh-vision","owner":"xiaoshihou514","short":"dsh-vision","slug":"xiaoshihou514/dsh-vision","desc":"极简的原生视觉能力增强，支持免费的智谱模型或本地千问视觉模型。","cmd":"dsh plugin --profile web add github:xiaoshihou514/dsh-vision","href":"/zh/p/xiaoshihou514/dsh-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":2,"added":"2026-08-14","npm":0,"name":"dsh-vision-bridge","owner":"ximengxiaolan","short":"dsh-vision-bridge","slug":"ximengxiaolan/dsh-vision-bridge","desc":"输入框贴图自动识别：由 OpenAI 兼容视觉模型转成文字描述后，再交给纯文本 DeepSeek 模型处理。","cmd":"dsh plugin --profile web add github:ximengxiaolan/dsh-vision-bridge","href":"/zh/p/ximengxiaolan/dsh-vision-bridge/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":2,"added":"2026-08-16","npm":0,"name":"dsh-image-vision","owner":"xsoc1","short":"dsh-image-vision","slug":"xsoc1/dsh-image-vision","desc":"纯文本 DeepSeek 的聊天识图插件：view_image 工具转发任意 OpenAI 兼容 VLM（本地 Ollama 或云端），对话框粘贴/拖拽的图片自动改写为 view_image 路径标记供模型查看。","cmd":"dsh plugin --profile web add github:xsoc1/dsh-image-vision","href":"/zh/p/xsoc1/dsh-image-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":1,"added":"2026-08-21","npm":0,"name":"dsh-client-vision#tool-vision","owner":"ankye","short":"dsh-client-vision#tool-vision","slug":"ankye/dsh-client-vision--packages-tool-vision","desc":"屏幕截图与外部视觉识别：take_screenshot、list_windows、analyze_image、view_image 四个工具，可配置 GPT 视觉通道（gpt-5.5 / gpt-5.6-sol / gpt-5.6-terra），API Key 经凭据服务存储，带设置卡片；view_image 在 Web 对话中显示截图，模型上下文只保留文字。","cmd":"dsh plugin --profile web add github:ankye/dsh-client-vision#path:/packages/tool-vision","href":"/zh/p/ankye/dsh-client-vision--packages-tool-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":1,"added":"2026-10-17","npm":0,"name":"local-ocr","owner":"Daive1119","short":"local-ocr","slug":"Daive1119/local-ocr","desc":"为纯文本模型注册一个 `ocr` 工具：调用仓库自带的离线脚本识别单张图片或 PDF 并返回文字——Windows 原生 OCR 优先，其次 RapidOCR，最后 Tesseract；识别过程在本机完成，不发网络请求、不需要 API key。需自备 PATH 上的 Python 3.11 并 `pip install winocr pillow`（winocr 为首选引擎，RapidOCR 与 Tesseract 为兜底）。图片返回逐行结果，PDF 返回逐页结果。","cmd":"dsh plugin --profile web add github:Daive1119/local-ocr","href":"/zh/p/Daive1119/local-ocr/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":1,"added":"2026-08-15","npm":0,"name":"dsh-plugin-mm-vision","owner":"Elohia","short":"dsh-plugin-mm-vision","slug":"Elohia/dsh-plugin-mm-vision","desc":"通感编码器（Synesthesia Encoder）：调用视觉模型把图片翻译成紧凑的结构化空间文字（画布/元素/百分比坐标），让纯文本 LLM 获得像素级看图能力（`mm_vision` 工具）。","cmd":"dsh plugin --profile web add github:Elohia/dsh-plugin-mm-vision","href":"/zh/p/Elohia/dsh-plugin-mm-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":1,"added":"2026-10-20","npm":0,"name":"dsh-vision-better","owner":"ethanweave","short":"dsh-vision-better","slug":"ethanweave/dsh-vision-better","desc":"为纯文本模型提供的图像识别转发：后台调用你配置的 GLM 兼容视觉接口分析图片，原模型在同一会话继续作答，你的消息、图片、模型与历史全程不变。需先运行仓库自带的 patches/apply-core-patch.cmd 给 DSH 打核心补丁，否则带图消息会在入库前被原版 DSH 拒绝（可用 restore 脚本还原）。","cmd":"dsh plugin --profile web add github:ethanweave/dsh-vision-better","href":"/zh/p/ethanweave/dsh-vision-better/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":1,"added":"2026-08-15","npm":0,"name":"dsh-tool-vision","owner":"gloryxpnv","short":"dsh-tool-vision","slug":"gloryxpnv/dsh-tool-vision","desc":"本地优先的结构化视觉插件：图片交给本地 OpenAI 兼容 VLM，返回 JSON 证据（摘要、逐字 OCR、版面区域、实体/关系、配色、显式不确定项），带反幻觉回退与可选粘贴/上传桥；零云端成本，图片不出本机。","cmd":"dsh plugin --profile web add github:gloryxpnv/dsh-tool-vision","href":"/zh/p/gloryxpnv/dsh-tool-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":1,"added":"2026-08-23","npm":0,"name":"dsh-vision-analysis","owner":"Harvey-Will","short":"dsh-vision-analysis","slug":"Harvey-Will/dsh-vision-analysis","desc":"DeepSeek Harness 图像理解插件：8 种分析模式（描述、OCR、图表取数、UI 评审、目标检测、对比、代码生成、诊断），支持任意 OpenAI/Anthropic 兼容视觉 API，内置免费视觉模型与限流自动切换。","cmd":"dsh plugin --profile web add github:Harvey-Will/dsh-vision-analysis","href":"/zh/p/Harvey-Will/dsh-vision-analysis/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":1,"added":"2026-08-25","npm":0,"name":"dsh-mmroute","owner":"jmxsxwyzjdwl","short":"dsh-mmroute","slug":"jmxsxwyzjdwl/dsh-mmroute","desc":"纯文本模型的透明多模态路由：每一次模型调用里的每张图片先由你自己配置的多模态理解模型全量转述（逐字 OCR、数据、不确定区、防注入），vision_relook 定点复看，图片类报错自动转述重试。不内置端点、不借登录态。","cmd":"dsh plugin --profile web add github:jmxsxwyzjdwl/dsh-mmroute","href":"/zh/p/jmxsxwyzjdwl/dsh-mmroute/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":1,"added":"2026-08-29","npm":0,"name":"deepseek-plugin","owner":"klingai-dev","short":"deepseek-plugin","slug":"klingai-dev/deepseek-plugin","desc":"一句话，让灵感从想法变成大片。在 DeepSeek Harness 中直接用自然语言调用可灵 AI，支持文生图、图生图、文生视频、图生视频、参考图创作、任务跟踪和结果预览。","cmd":"dsh plugin --profile web add github:klingai-dev/deepseek-plugin","href":"/zh/p/klingai-dev/deepseek-plugin/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":1,"added":"2026-08-17","npm":0,"name":"dsh-eyes","owner":"Leeminjing","short":"dsh-eyes","slug":"Leeminjing/dsh-eyes","desc":"为纯文本 DeepSeek 模型提供按需视觉：上传图片后，模型通过 view_image 工具调用任意 OpenAI 兼容视觉端点（默认 Qwen/DashScope）。","cmd":"dsh plugin --profile web add github:Leeminjing/dsh-eyes","href":"/zh/p/Leeminjing/dsh-eyes/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":1,"added":"2026-10-17","npm":0,"name":"godot-imagegen","owner":"LINinLIN-0079","short":"godot-imagegen","slug":"LINinLIN-0079/godot-imagegen","desc":"本机美术流水线：把本机已装的 SD WebUI 与 ComfyUI 接进 DSH——10 个 image_* 工具 + 「美术工坊」侧边栏面板（内嵌 ComfyUI 画布），用整合包自带的 Python 做抠图 / 放大 / 无缝平铺后处理；填写 Godot 项目根目录后，产物直接落成带 res:// 路径的游戏资产。","cmd":"dsh plugin --profile web add github:LINinLIN-0079/godot-imagegen","href":"/zh/p/LINinLIN-0079/godot-imagegen/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":1,"added":"2026-08-15","npm":0,"name":"dsh-llm-deepseek-vision","owner":"NagasakiSoyo-ui","short":"dsh-llm-deepseek-vision","slug":"NagasakiSoyo-ui/dsh-llm-deepseek-vision","desc":"视觉增强的 DeepSeek 适配器：由视觉模型把图片描述成文字，纯文本 DeepSeek 模型再基于描述推理。","cmd":"dsh plugin --profile web add github:NagasakiSoyo-ui/dsh-llm-deepseek-vision","href":"/zh/p/NagasakiSoyo-ui/dsh-llm-deepseek-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":1,"added":"2026-08-18","npm":0,"name":"dsh-tool-vision","owner":"wanshichenguang","short":"dsh-tool-vision","slug":"wanshichenguang/dsh-tool-vision","desc":"面向模型的 image_describe（识图）工具（DashScope OpenAI 兼容接口，qwen3.7-flash）+ 粘贴桥：纯文本模型会话里粘贴的图片在发送时自动转为文件路径并在聊天记录中回显，不再触发图片准入拦截。自带 DASHSCOPE_API_KEY；端点/模型/预算可配置，重定向防护 HTTP 客户端，所有 Agent 预设均可用。","cmd":"dsh plugin --profile web add github:wanshichenguang/dsh-tool-vision","href":"/zh/p/wanshichenguang/dsh-tool-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-08-27","npm":0,"name":"dsh-vision-bridge","owner":"alaxrpg","short":"dsh-vision-bridge","slug":"alaxrpg/dsh-vision-bridge","desc":"通过已配置的 DSH Provider 或 OpenAI 兼容端点提供图片输入与识别。","cmd":"dsh plugin --profile web add github:alaxrpg/dsh-vision-bridge","href":"/zh/p/alaxrpg/dsh-vision-bridge/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-09-03","npm":0,"name":"dsh-image-generation#tool-image-generation","owner":"ankye","short":"dsh-image-generation#tool-image-generation","slug":"ankye/dsh-image-generation--packages-tool-image-generation","desc":"面向模型的图像生成工具，提供可配置通道和标准化图像参数。","cmd":"dsh plugin --profile web add github:ankye/dsh-image-generation#path:/packages/tool-image-generation","href":"/zh/p/ankye/dsh-image-generation--packages-tool-image-generation/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-10-16","npm":0,"name":"opencode-eyes","owner":"bauerelizabeth07139","short":"opencode-eyes","slug":"bauerelizabeth07139/opencode-eyes","desc":"给不具备多模态能力的模型一双眼睛:把随插件分发的 stdio MCP 服务器挂成 mcp__opencode_eyes__describe_image,把图片交给 StepFun step-3.7-flash 并返回详细描述。","cmd":"dsh plugin --profile web add github:bauerelizabeth07139/opencode-eyes","href":"/zh/p/bauerelizabeth07139/opencode-eyes/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-10-16","npm":0,"name":"opencode-eyes-nvidia","owner":"bauerelizabeth07139","short":"opencode-eyes-nvidia","slug":"bauerelizabeth07139/opencode-eyes-nvidia","desc":"走 NVIDIA NIM 的视觉能力:把随插件分发的 stdio MCP 服务器挂成 mcp__opencode_eyes_nvidia__describe_image,用 7 个托管视觉模型之一(默认 MiniMax-M3)描述图片,并支持多 Key 轮换。","cmd":"dsh plugin --profile web add github:bauerelizabeth07139/opencode-eyes-nvidia","href":"/zh/p/bauerelizabeth07139/opencode-eyes-nvidia/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-10-16","npm":0,"name":"opencode-ppocr-mcp","owner":"bauerelizabeth07139","short":"opencode-ppocr-mcp","slug":"bauerelizabeth07139/opencode-ppocr-mcp","desc":"本地 OCR,不把文件发到云端:把 PP-OCRv6 Medium stdio MCP 服务器(PaddleOCR + ONNX Runtime CPU)挂成 mcp__ppocr__ocr_image 与 mcp__ppocr__ocr_pdf,识别图片和多页 PDF。","cmd":"dsh plugin --profile web add github:bauerelizabeth07139/opencode-ppocr-mcp","href":"/zh/p/bauerelizabeth07139/opencode-ppocr-mcp/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-08-17","npm":0,"name":"dsh-vision-solution","owner":"br1nosense","short":"dsh-vision-solution","slug":"br1nosense/dsh-vision-solution","desc":"为 DSH 纯文本模型补充视觉能力：识图技能（图片理解/OCR/文档解析，竞速池→自定义通道→本地）+ 幂等宿主补丁，让图片消息能送达模型侧。","cmd":"dsh plugin --profile web add github:br1nosense/dsh-vision-solution","href":"/zh/p/br1nosense/dsh-vision-solution/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-10-17","npm":0,"name":"dsh-image-compare","owner":"ChordXD","short":"dsh-image-compare","slug":"ChordXD/dsh-image-compare","desc":"在右栏新增一个面板，用来并排对比用户自己从工作区挑选的一组图片——每页 1/2/3/4/6 张、每张独立缩放与平移、可把某一张的缩放与显示位置同步到整组、缩略图可拖动排序。图片由浏览器直接加载，不进入模型。","cmd":"dsh plugin --profile web add github:ChordXD/dsh-image-compare","href":"/zh/p/ChordXD/dsh-image-compare/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-10-17","npm":0,"name":"dsh-vision-diff","owner":"dingchenhui0618-arch","short":"dsh-vision-diff","slug":"dingchenhui0618-arch/dsh-vision-diff","desc":"面向视觉回归的像素级截图比对。输入两张本地 PNG，第一行是可机器判定的结论（`PIXEL_CHANGED=true|false`，可以直接 grep 当断言），随后是差异比例、差异区域包围盒、最大通道差与平均绝对误差，并生成一份自包含的 HTML 对照页。可用绝对或比例坐标声明忽略区，排除时钟、光标这类每次渲染都会变的部分。它刻意不做内容识别——只回答「变没变」，因为这一步可以做成确定性的。全部本地完成：不调模型、不联网、不上传。","cmd":"dsh plugin --profile web add github:dingchenhui0618-arch/dsh-vision-diff","href":"/zh/p/dingchenhui0618-arch/dsh-vision-diff/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-08-19","npm":0,"name":"dsh-maclens","owner":"Harzva","short":"dsh-maclens","slug":"Harzva/dsh-maclens","desc":"苹果设备端 Vision 工具：本地 OCR（含中文）、图像分类、人脸检测、文档版面分析与综合描述，100% 离线、无需 API key，支持长截图切片。","cmd":"dsh plugin --profile web add github:Harzva/dsh-maclens","href":"/zh/p/Harzva/dsh-maclens/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-08-18","npm":0,"name":"wechat-ocr","owner":"hawkhai","short":"wechat-ocr","slug":"hawkhai/wechat-ocr","desc":"DSH 的本地微信 OCR 工具：`wechat_ocr_recognize` 针对本地图片路径返回识别文本和引擎的结构化结果。","cmd":"dsh plugin --profile web add github:hawkhai/wechat-ocr","href":"/zh/p/hawkhai/wechat-ocr/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-08-18","npm":0,"name":"win11-oneocr","owner":"hawkhai","short":"win11-oneocr","slug":"hawkhai/win11-oneocr","desc":"DSH 的本地 Windows 11 OneOCR 工具：`oneocr_recognize` 返回 OCR 文本，以及包含行/词多边形、置信度、旋转角度和手写体样式的结构化结果。","cmd":"dsh plugin --profile web add github:hawkhai/win11-oneocr","href":"/zh/p/hawkhai/win11-oneocr/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-08-29","npm":0,"name":"dsh-auto-vision","owner":"k2d5rqjpkg-art","short":"dsh-auto-vision","slug":"k2d5rqjpkg-art/dsh-auto-vision","desc":"按需自动把 DeepSeek 路由切到视觉模型：flash 主会话直接切（A），pro 保持深推理并交给视觉子代理读图（B），子代理恒切，含失败回退，无需手动切模型。","cmd":"dsh plugin --profile web add github:k2d5rqjpkg-art/dsh-auto-vision","href":"/zh/p/k2d5rqjpkg-art/dsh-auto-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-08-17","npm":0,"name":"dsh-vision-plugin","owner":"ld-1101","short":"dsh-vision-plugin","slug":"ld-1101/dsh-vision-plugin","desc":"让纯文本模型也能看图：发图自动用视觉模型描述（默认提示词），描述不足时对话模型自动生成更具体的提示词重新解析；系统模型/自定义双模式 + GUI 可视化配置、Key 脱敏安全、针对 DSH 0.1.0-rc.6 的宿主小补丁（见仓库 README）。","cmd":"dsh plugin --profile web add github:ld-1101/dsh-vision-plugin","href":"/zh/p/ld-1101/dsh-vision-plugin/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-10-18","npm":0,"name":"dsh-image-insight","owner":"Monter-code","short":"dsh-image-insight","slug":"Monter-code/dsh-image-insight","desc":"结构化图像识别：本地 sharp 感知 + 云端视觉模型，一次调用拿到描述、带坐标的 OCR、 目标定位与细粒度属性；图片指纹缓存让重复图片毫秒返回且不消耗模型额度。","cmd":"dsh plugin --profile web add github:Monter-code/dsh-image-insight","href":"/zh/p/Monter-code/dsh-image-insight/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-08-19","npm":0,"name":"dsh-omni-vision","owner":"Renji004","short":"dsh-omni-vision","slug":"Renji004/dsh-omni-vision","desc":"为纯文本模型装上本地眼睛：eyes_render 在 Web 界面画布上绘制文字/图形/Mermaid，eyes_paste 接收用户粘贴的图片，eyes_ocr 用 Windows 自带 OCR 离线读文字，eyes_analyze 把像素解析成结构化数据，全程无需视觉模型。","cmd":"dsh plugin --profile web add github:Renji004/dsh-omni-vision","href":"/zh/p/Renji004/dsh-omni-vision/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-08-16","npm":0,"name":"dsh-easyvision","owner":"s3yf1337","short":"dsh-easyvision","slug":"s3yf1337/dsh-easyvision","desc":"让纯文本模型拥有视觉：describe_image 工具调用 dsh 模型列表中的视觉模型描述图片，走 harness 自身的 LLM 运行时。","cmd":"dsh plugin --profile web add github:s3yf1337/dsh-easyvision","href":"/zh/p/s3yf1337/dsh-easyvision/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-10-20","npm":0,"name":"dsh-video-kit","owner":"shisuiqingqian","short":"dsh-video-kit","slug":"shisuiqingqian/dsh-video-kit","desc":"用 ffmpeg 与 Paraformer 读取本地视频或网页链接，返回带时间轴的逐字稿与可逐帧打开的关键帧图片；可按关键词检索候选视频，首次使用时自动安装运行时依赖。需 PATH 上有 Python 3.8+；video_discover 还需一次性登录抖音/小红书。","cmd":"dsh plugin --profile web add \"https://github.com/shisuiqingqian/dsh-video-kit/releases/latest/download/dsh-video-kit.tgz\"","href":"/zh/p/shisuiqingqian/dsh-video-kit/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-10-17","npm":0,"name":"dsh-paddle-ocr","owner":"tenglongbohai","short":"dsh-paddle-ocr","slug":"tenglongbohai/dsh-paddle-ocr","desc":"为纯文本模型添加 PaddleOCR 视觉能力，支持 Agnes、DeepSeek、MiniMax、GLM、Mimo，支持本地文件与 URL。每日 2 万页免费额度；需先申请 PaddleOCR API Token（免费，审核约 1-3 个工作日）。","cmd":"dsh plugin --profile web add github:tenglongbohai/dsh-paddle-ocr","href":"/zh/p/tenglongbohai/dsh-paddle-ocr/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-10-17","npm":0,"name":"dsh-vision-unlock","owner":"wmk66ds","short":"dsh-vision-unlock","slug":"wmk66ds/dsh-vision-unlock","desc":"移除拦截图片请求的多模态能力门禁：改写 llm.resolveModelInfo 的返回值与适配器模型表，为未声明 input 的路由补上 image 输入能力，使其可以发送和读取图片。不改 app.asar，也不改配置树。","cmd":"dsh plugin --profile web add github:wmk66ds/dsh-vision-unlock","href":"/zh/p/wmk66ds/dsh-vision-unlock/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-10-17","npm":0,"name":"dsh-omni-modal","owner":"yzhclcl","short":"dsh-omni-modal","slug":"yzhclcl/dsh-omni-modal","desc":"为 DeepSeek Harness 增加「全模态设置」分区：可把生图 / 生视频指向任意 OpenAI 兼容端点，多套配置档案可随时切换，图片与视频各自独立的接口地址、API Key、模型、路径与尺寸，支持自由填写额外请求参数、不消耗额度的连通性测试，并注册 generate_image / generate_video / check_video 三个模型可见工具，异步视频任务可跨重启保留。","cmd":"dsh plugin --profile web add github:yzhclcl/dsh-omni-modal","href":"/zh/p/yzhclcl/dsh-omni-modal/"},{"cat":"vision","tag":"视觉与多模态","dl":null,"stars":0,"added":"2026-10-20","npm":0,"name":"dsh-ocr","owner":"zirscard","short":"dsh-ocr","slug":"zirscard/dsh-ocr","desc":"为 DeepSeek Harness 注册两个工具：`doc_ocr` 把扫描件 PDF 与图片转成结构化文本（逐页文本、带资产负债表平衡勾稽的财务报表行、合同条款命中），可用本机 RapidOCR 或智谱 GLM-OCR 云端引擎；`doc_preprocess` 用 7 阶段管线把一批原始文件整理成 Obsidian 可读知识库，含后端路由与 sha256 增量跳过。勾稽不平的数字只报出、不编造。本机引擎需 Python 3 并安装 rapidocr-onnxruntime、pypdfium2、Pillow（pip install -r requirements.txt）；GLM 引擎需设置 ZHIPU_API_KEY。","cmd":"dsh plugin --profile web add \"https://github.com/zirscard/dsh-ocr/releases/download/v0.3.0/dsh-ocr.tgz\"","href":"/zh/p/zirscard/dsh-ocr/"}]}