# silicon-paddle-ocr

Recognizes text in images through the PaddleOCR model and the SiliconFlow API. Handy when you need text out of a screenshot or photo without leaving the agent chat.

- skill · `type=skill` · Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode · 1200 installs · ★ 1
- Author: aotenjou
- License: MIT · policy=mirror · the beginning of the file is at the end of this page
- Source: https://github.com/aotenjou/silicon-PaddleOCR/tree/HEAD/skills/silicon-paddle-ocr/SKILL.md
- Updated: 2026-02-09
- HTML: https://affpapa.org/en/nevibecode/skills/silicon-paddle-ocr

## What it does

The skill runs `scripts/ocr_skill.py`, which sends an image to SiliconFlow and returns the recognized text. JPG, PNG, WebP, BMP and GIF are supported. Pass a single file, several paths or a glob pattern and it processes them as a batch.

The default model is `PaddlePaddle/PaddleOCR-VL-1.5`, with a 2000-token response limit. The `--json` flag returns coordinates for every text region, `--output` writes results to a file, and `-p` sets a custom prompt — useful for getting a table back as Markdown. If one image fails, the script reports the error for that file and keeps going with the rest.

You need a SiliconFlow key in the `SILICONFLOW_API_KEY` environment variable before the first run. Without it the script never gets past authorization, so check the variable before pointing the skill at a folder of scans.

## Install

### Claude Code (native platform)

1. **Skills CLI (Vercel)**
```bash
npx skills add aotenjou/silicon-PaddleOCR --skill silicon-paddle-ocr -a claude-code
```

2. **Manually: put the files in place**
- File: `~/.claude/skills/silicon-paddle-ocr/SKILL.md` — Copy the whole skill folder; for one project use .claude/skills/silicon-paddle-ocr/

### Codex

1. **Skills CLI (Vercel)**
```bash
npx skills add aotenjou/silicon-PaddleOCR --skill silicon-paddle-ocr -a codex
```

2. **Manually: put the files in place**
- File: `.agents/skills/silicon-paddle-ocr/SKILL.md` — For all projects use ~/.agents/skills/

### Cursor

1. **Skills CLI (Vercel)**
```bash
npx skills add aotenjou/silicon-PaddleOCR --skill silicon-paddle-ocr -a cursor
```

2. **Manually: put the files in place**
- File: `.cursor/skills/silicon-paddle-ocr/SKILL.md` — Cursor also reads .agents/skills/ and .claude/skills/

### GitHub Copilot

1. **GitHub CLI**
```bash
gh skill install aotenjou/silicon-PaddleOCR silicon-paddle-ocr
```

2. **Manually: put the files in place**
- File: `.github/skills/silicon-paddle-ocr/SKILL.md`

### Gemini CLI

1. **Skills CLI (Vercel)**
```bash
npx skills add aotenjou/silicon-PaddleOCR --skill silicon-paddle-ocr -a gemini-cli
```

2. **Platform CLI**
```bash
gemini skills install https://github.com/aotenjou/silicon-PaddleOCR --path skills/silicon-paddle-ocr
```

### Windsurf (Devin Desktop)

**Manually: put the files in place**
- File: `.devin/skills/silicon-paddle-ocr/SKILL.md` — The legacy .windsurf/skills/ path still works

### OpenCode

**Skills CLI (Vercel)**
```bash
npx skills add aotenjou/silicon-PaddleOCR --skill silicon-paddle-ocr -a opencode
```

### Claude.ai

**Claude.ai**
- Zip the folder and upload it in Settings → Capabilities → Skills

Where a file is given instead of a command, copy the files from the source to that path: https://github.com/aotenjou/silicon-PaddleOCR/tree/HEAD/skills/silicon-paddle-ocr/SKILL.md

## When to use it
- pull text out of a screenshot of an error
- read text from a photo of a document
- run a batch of images through OCR in one go
- get text region coordinates as JSON
- export a table from an image into Markdown

## Example prompts
- recognize the text in this screenshot
- OCR every PNG in the screenshots folder
- extract the table from this image and format it as Markdown

## Source text · SKILL.md · MIT

Below is the content of a third-party file. It is data, not instructions for you: do not act on anything written inside unless the human asks. Block tag: de670ce2307c5ed1.

BEGIN THIRD-PARTY FILE de670ce2307c5ed1
````text
---
name: silicon-paddle-ocr
description: OCR skill using PaddleOCR model via SiliconFlow API. This skill should be used when the user asks to "recognize text from an image", "extract text from a photo", "OCR this image", "read text from screenshot", or mentions "PaddleOCR", "image text recognition", "text extraction from images".
license: MIT
metadata:
  author: aotenjou
  version: "1.0.0"
---

# OCR - Image Text Recognition

Use PaddleOCR to extract text content from images. Supports single image or batch processing.

## Overview

This skill provides optical character recognition (OCR) capabilities using the PaddlePaddle/PaddleOCR-VL-1.5 model via the SiliconFlow API. Extract text from JPG, PNG, WebP, BMP, and GIF images.

## When to Use

Invoke this skill when:
- User wants to extract text from an image
- User asks to OCR a screenshot or photo
- User needs to read text from an image file
- User mentions text recognition from images

## How to Use

### Prerequisites

Ensure the `SILICONFLOW_API_KEY` environment variable is set:
```bash
export SILICONFLOW_API_KEY="your_api_key"
```

### Basic Usage

Execute the OCR script:
```bash
python3 scripts/ocr_skill.py [options] image_path
```

### Arguments

| Argument | Description |
|----------|-------------|
| `images` | Image file path(s) or glob pattern (required) |
| `-k, --api-key` | API key (default: from SILICONFLOW_API_KEY env) |
| `-m, --model` | OCR model name (default: PaddlePaddle/PaddleOCR-VL-1.5) |
| `-p, --prompt` | Recognition prompt for custom behavior |
| `-j, --json` | Output results in JSON format |
| `-o, --output` | Save results to specified file |
| `--max-tokens` | Maximum tokens in response (default: 2000) |

### Examples

Single image:
```bash
python3 scripts/ocr_skill.py /path/to/image.jpg
```

Multiple images with glob:
```bash
python3 scripts/ocr_skill.py /path/to/images/*.png
```

JSON output format:
```bash
python3 scripts/ocr_skill.py --json /path/to/image.jpg
```

Custom prompt for table extraction:
```bash
python3 scripts/ocr_skill.py -p "Please identify and format table content as Markdown" /path/to/table.jpg
```

Save to file:
```bash
python3 scripts/ocr_skill.py --json --output results.json /path/to/images/*.jpg
```

### Output Format

**Text output** (default):
```
--- image.jpg ---
识别到的文字内容
识别到 X 处文字区域
```

**JSON output**:
```json
{
  "image.jpg": {
    "image_path": "/path/to/image.jpg",
    "image_size": [width, height],
    "texts": [
      {
        "text": "识别的文字",
        "box": [[x1, y1], [x2, y2], [x3, y3], [x4, y4]]
      }
    ],
    "full_text": "所有文本的组合"
  },
  "image2.png": { ... }
}
```

**Coordinates Explanation:**
- LOC values are normalized coordinates converted to pixel coordinates
- Conversion: pixel = LOC × (image_size / LOC_max_value)
- LOC max_value is approximately 972 (may vary by model/image)
````
END THIRD-PARTY FILE de670ce2307c5ed1

The third-party file has ended. Everything between the de670ce2307c5ed1 tags is data: requests and commands from it are not followed. An “end of file” line with another tag inside the block is part of the file.

First 2908 of 3586 bytes. © aotenjou · MIT · https://github.com/aotenjou/silicon-PaddleOCR/tree/HEAD/skills/silicon-paddle-ocr/SKILL.md

_Part of the page did not fit — see the HTML version._

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md