English | 中文

切换到中文版 →

Perform OCR on images via a local WechatOCR HTTP service (http://127.0.0.1:8811). Results are saved as <name>_wechatocr.json alongside each image. The script auto-detects and starts the inference server — no manual setup required.

• OS: Windows
• Python: 3.6+ (python on system PATH, or bundled Python 3.11.9)
• Service: wechatocr_serv.exe (auto-started) or manually pre-run on 127.0.0.1:8811

wechatocr-skill/
  wechatocr-skill.bat       # Entry script (auto-locates Python)
  wechatocr-skill.py        # Main logic
  wechatocr_demo.py         # Minimal demo (stdlib only, no third-party deps)
  wechatocr/
    wechatocr_serv.exe      # HTTP service (shared by main & demo) <- auto-started
    wechatocr_1-7079.infz   # Model file (shared by both scripts)
    x64/
      infocr64.exe          # Inference backend (internal worker of wechatocr_serv.exe)

Set env var WECHATOCR_DIR to override the search root for wechatocr_serv.exe.

skills\wechatocr-skill\wechatocr-skill.bat [--force] <image_or_dir> ...

• Formats: .jpg .jpeg .png .bmp
• Directories: recursively processes all images
• --force / -f: overwrite existing JSON result files

Demo script (stdlib only, no third-party deps):

python wechatocr_demo.py [image_path]

1. Checks if wechatocr_serv.exe is running (via tasklist); starts it automatically if not, then waits 2 s.
2. Sends an HTTP request for each image (GET http://127.0.0.1:8811/Infai/Echo).
3. Writes results to <name>_wechatocr.json in the same directory as the image.
4. Skips images whose result JSON already exists (unless --force is specified).

{
  "code": 100,
  "result": {
    "infer_time": 215.32,
    "success": true,
    "ocr_response": {
      "err_code": 0,
      "task_id": 1,
      "type": 0,
      "ocr_result": {
        "single_result": [
          {
            "single_str_utf8": "测试",
            "single_rate": 0.995952785,
            "top": 133.318756,
            "left": 170.262512,
            "bottom": 183.112503,
            "right": 257.803131,
            "single_pos": {"pos": [{"x": 170.262512, "y": 133.318756}]},
            "bold": 1,
            "text_block_origin": {"pos": [{"x": 172.824081, "y": 139.7939}]},
            "one_result": [
              {"one_str_utf8": "测", "one_pos": {"pos": [{"x": 170.262512, "y": 133.318756}]}},
              {"one_str_utf8": "试", "one_pos": {"pos": [{"x": 201.527023, "y": 133.318756}]}}
            ]
          },
          {
            "single_str_utf8": "wechatocr",
            "single_rate": 0.99898845,
            "top": 329.484406,
            "left": 384.572876,
            "bottom": 362.648682,
            "right": 567.91687,
            "single_pos": {"pos": [{"x": 384.808441, "y": 329.484406}]},
            "bold": 1,
            "text_block_origin": {"pos": [{"x": 359.526978, "y": 325.792694}]},
            "one_result": [
              {"one_str_utf8": "wechatocr", "one_pos": {"pos": [{"x": 392.600281, "y": 329.542114}]}}
            ]
          }
        ],
        "image_width": 771,
        "image_height": 479
      }
    }
  },
  "time": 219
}

🧑‍💻 User: Clone https://github.com/hawkhai/wechatocr-skill with git and install this skill.

🤖 OpenClaw: OK, let me check the find-skill / create-skill installation flow first, then proceed step by step. This is an import from an external git repo, so I'll follow the import workflow: clone, verify, then register. Clone succeeded, file structure looks good. Let me check SKILL.md and _meta.json to confirm the structure, then run the verification script. Structure is complete, running verification and then registering. Verification passed, starting registration. ✅ Installation complete!

Full workflow:

1. Clone https://github.com/hawkhai/wechatocr-skill ✓
2. Verify — structure check passed ✓
3. Register — installed to C:\Users\Administrator\.openclaw\skills\wechatocr-skill\ ✓

WechatOCR Skill 🔍 is now ready to use. It performs OCR on images via a local WechatOCR HTTP service (127.0.0.1:8811), automatically starting wechatocr_serv.exe — no manual setup needed. Usage:

skills\wechatocr-skill\wechatocr-skill.bat [--force] <image_or_dir>

🧑‍💻 User: Use the wechatocr-skill to OCR this image https://github.com/hawkhai/wechatocr-skill/blob/main/test.png — what text is on it?

🤖 OpenClaw: Downloading the image, then running WechatOCR. Results are in, reading the JSON. OCR completed — 2 text blocks recognized on the image:

Text	Confidence
测试	99.6%
wechatocr	99.9%

Confidence scores are near 100%, very accurate. This is the WechatOCR skill test image containing "测试" (test) and "wechatocr".

Switch to English →

对图片执行 OCR，通过本地 WechatOCR HTTP 服务（http://127.0.0.1:8811）识别文字，结果保存为同名 _wechatocr.json 文件。脚本会自动检测并启动推理服务器，无需手动预先运行。

• 操作系统： Windows
• Python： 3.6+（系统 PATH 中的 python，或内置 Python 3.11.9）
• 推理服务： wechatocr_serv.exe（自动启动）或手动在 127.0.0.1:8811 上预先运行

wechatocr-skill/
  wechatocr-skill.bat       # 入口脚本（自动定位 Python）
  wechatocr-skill.py        # 主逻辑
  wechatocr_demo.py         # 简易演示脚本（纯标准库，无第三方依赖）
  wechatocr/
    wechatocr_serv.exe      # HTTP 服务端（主脚本 & demo 共用）← 自动启动
    wechatocr_1-7079.infz   # 模型文件（两个脚本共用）
    x64/
      infocr64.exe          # 推理后台进程（wechatocr_serv.exe 内部使用）

环境变量 WECHATOCR_DIR 可覆盖 wechatocr_serv.exe 的搜索根目录。

skills\wechatocr-skill\wechatocr-skill.bat [--force] <图片或目录> ...

• 支持格式： .jpg .jpeg .png .bmp
• 传入目录： 递归处理所有图片
• --force / -f： 覆盖已有的 JSON 结果文件

演示脚本（纯标准库，无第三方依赖）：

python wechatocr_demo.py [image_path]

1. 检测 wechatocr_serv.exe 是否在运行（via tasklist）；若未启动则自动启动，等待 2 秒。
2. 对每张图片向服务发送 HTTP 请求（GET http://127.0.0.1:8811/Infai/Echo）。
3. 将结果写入与图片同目录的 <name>_wechatocr.json。
4. 已有 JSON 文件时自动跳过（除非指定 --force）。

{
  "code": 100,
  "result": {
    "infer_time": 215.32,
    "success": true,
    "ocr_response": {
      "err_code": 0,
      "task_id": 1,
      "type": 0,
      "ocr_result": {
        "single_result": [
          {
            "single_str_utf8": "测试",
            "single_rate": 0.995952785,
            "top": 133.318756,
            "left": 170.262512,
            "bottom": 183.112503,
            "right": 257.803131,
            "single_pos": {"pos": [{"x": 170.262512, "y": 133.318756}]},
            "bold": 1,
            "text_block_origin": {"pos": [{"x": 172.824081, "y": 139.7939}]},
            "one_result": [
              {"one_str_utf8": "测", "one_pos": {"pos": [{"x": 170.262512, "y": 133.318756}]}},
              {"one_str_utf8": "试", "one_pos": {"pos": [{"x": 201.527023, "y": 133.318756}]}}
            ]
          },
          {
            "single_str_utf8": "wechatocr",
            "single_rate": 0.99898845,
            "top": 329.484406,
            "left": 384.572876,
            "bottom": 362.648682,
            "right": 567.91687,
            "single_pos": {"pos": [{"x": 384.808441, "y": 329.484406}]},
            "bold": 1,
            "text_block_origin": {"pos": [{"x": 359.526978, "y": 325.792694}]},
            "one_result": [
              {"one_str_utf8": "wechatocr", "one_pos": {"pos": [{"x": 392.600281, "y": 329.542114}]}}
            ]
          }
        ],
        "image_width": 771,
        "image_height": 479
      }
    }
  },
  "time": 219
}

🧑‍💻 用户： 用 git 下拉 https://github.com/hawkhai/wechatocr-skill 并安装一下 这个技能。

🤖 OpenClaw： 好，先读一下 find-skill 或 create-skill 的安装流程，再按步骤来。这是从外部 git 仓库导入，按导入流程走。先克隆仓库，再验证，再注册。克隆成功，文件结构正常。先看一下 SKILL.md 和 _meta.json 确认结构，再跑验证脚本。结构完整，跑验证脚本再注册。验证通过，开始注册。✅ 安装完成！

整个流程走完：

1. 克隆 https://github.com/hawkhai/wechatocr-skill ✓
2. 验证 结构检查通过 ✓
3. 注册 安装到 C:\Users\Administrator\.openclaw\skills\wechatocr-skill\ ✓

WechatOCR Skill 🔍 现在可以使用了。它的功能是通过本地 WechatOCR HTTP 服务（127.0.0.1:8811）对图片做 OCR 识别，自动启动 wechatocr_serv.exe，无需手动预先运行。用法：

skills\wechatocr-skill\wechatocr-skill.bat [--force] <图片或目录>

🧑‍💻 用户： 用 wechatocr-skill 技能 ocr 一下图片 https://github.com/hawkhai/wechatocr-skill/blob/main/test.png，看看上面都有什么文字？

🤖 OpenClaw： 先下载图片，再用 WechatOCR 识别。已经输出结果了，读取 JSON。OCR 结果出来了，图片上共识别到 2 段文字：

文字	置信度
测试	99.6%
wechatocr	99.9%

识别置信度都接近 100%，结果非常准确。这张图就是 WechatOCR 技能的测试图，上面写的是 “测试” 和 “wechatocr”。