ComfyUI-PaddleOCR-VL

A ComfyUI custom node that integrates PaddleOCR for powerful and accurate text detection and recognition.

This node leverages the PaddlePaddle deep learning framework to provide industry-leading OCR capabilities directly within your ComfyUI workflows. It supports multiple languages and offers high accuracy for various scene text recognition tasks.

Features

• Text Detection & Recognition: Extract text from images with high precision.
• Multilingual Support: Supports Chinese, English, Japanese, Korean, French, German, and more.
• Model Version Selection: Choose between PP-OCRv5, PP-OCRv4, and PP-OCRv3 models.
• Auto-Orientation: Automatically detects and corrects text orientation (e.g., vertical text).

Installation

1. Navigate to your ComfyUI custom nodes directory:

   cd ComfyUI/custom_nodes

2. Clone this repository:

   git clone git@github.com:kaili-yang/ComfyUI-PaddleOCR-VL.git

3. Install the required dependencies:

   pip install paddlepaddle paddleocr opencv-python-headless

4. Restart ComfyUI.

Usage

1. Add Node: Right-click in the ComfyUI canvas and search for PaddleOCR Text Detection. You can typically find it under the PaddleOCR category.
2. Connect Input: Connect an image source (e.g., Load Image) to the image input of the PaddleOCR node.
3. Configure Parameters:
   • language: Select the language of the text in the image (default: ch for Chinese).
   • ocr_version: Choose the OCR model version (e.g., PP-OCRv5).
   • vertical_direction: Enable this if the text might be vertical or rotated.
4. Get Output: The node outputs a text string containing all recognized text from the image. You can connect this to a Show Text node or use it in other text processing workflows.

Credits

This project wraps the amazing PaddleOCR library by PaddlePaddle.

License

Apache 2.0