ClassifAI Image Text Extraction Data Flow (with OpenAI)

This diagram outlines the sequence of events when ClassifAI's Image Text Extraction feature processes an image to extract text, using OpenAI (e.g., a vision-capable GPT model) as the configured AI provider. This flow can be initiated manually from the Media Modal or attachment edit screen, or automatically upon image upload.

Automatic Generation on Upload

Text can also be extracted automatically when an image is uploaded:

1. User uploads an image.
2. WordPress core triggers the wp_generate_attachment_metadata hook.
3. ImageTextExtraction::generate_ocr_text() is called.
4. This method internally calls its run() method (which then calls the ChatGPT_Provider_PHP similar to the manual flow) to get the text.
5. The text is then saved using the save() method to the attachment's post_content.

Layers Involved

• WordPress Application Layer:
  • User: The end-user interacting with the WordPress Media Library or editor.
  • WordPress Admin UI (Media Modal/Attachment Editor): The interface for managing media.
  • ClassifAI Admin JS: JavaScript handling client-side interaction for text extraction.
  • WordPress REST API: The /wp-json/ interface, including ClassifAI's custom endpoint.
  • ClassifAI ImageTextExtraction Class (ImageTextExtraction_PHP): The PHP class (ImageTextExtraction.php) containing the server-side logic for this feature.
  • ClassifAI OpenAI ChatGPT Provider Class (ChatGPT_Provider_PHP): The PHP class (ChatGPT.php) responsible for communicating with the OpenAI API.
• Database Layer:
  • WordPress Database (WP_DB):
    • wp_posts: Stores attachment details (post_type 'attachment') and the extracted text in the post_content field of the attachment.
    • wp_options: Stores ClassifAI plugin settings, including the image text extraction prompt and OpenAI API key (e.g., under classifai_feature_image_to_text_generator option).
• API Layer:
  • WordPress REST API (Internal): Endpoint /wp-json/classifai/v1/ocr/{attachment_id}.
  • OpenAI GPT API (External): The AI service endpoint (e.g., GPT-4 Vision).
• AI Provider:
  • OpenAI GPT API: The specific AI model service (e.g., a vision-capable GPT model) used for analyzing the image and extracting text based on the provided prompt.

Data Flow Summary

1. User Action (Manual): The user initiates image text extraction for an image via the WordPress Admin UI (e.g., Media Modal by clicking "Scan image for text" or via the attachment edit screen).
2. Client-Side Request: JavaScript makes a GET request to the ClassifAI REST API endpoint /wp-json/classifai/v1/ocr/{attachment_id}, passing the attachment ID.
3. Server-Side Processing (ClassifAI - ImageTextExtraction_PHP):
   • The ImageTextExtraction.php class handles the request.
   • It performs permission checks (user capability, feature enabled).
   • It retrieves the image URL using wp_get_attachment_url().
   • It fetches the configured prompt for text extraction and OpenAI API key from wp_options (via classifai_feature_image_to_text_generator setting).
4. AI Provider Request (ClassifAI - ChatGPT_Provider_PHP):
   • The ChatGPT.php provider class receives the image URL and the specific prompt for extracting text.
   • It sends the image URL and the prompt to the OpenAI GPT API (a vision-capable model).
5. AI Provider Response: OpenAI processes the request and returns the extracted text.
6. Server-Side Response & Save (ClassifAI - ImageTextExtraction_PHP):
   • The provider class returns the extracted text string.
   • The ImageTextExtraction.php class receives this string.
   • The save() method is called, which uses wp_update_post() to save the extracted text into the post_content field of the attachment in the wp_posts table.
   • The ClassifAI REST endpoint sends the extracted text (or a success message) back to the client.
7. Client-Side Display: JavaScript displays a notification or the extracted text (if applicable to the UI context).
8. Automatic Flow: Alternatively, on image upload, the wp_generate_attachment_metadata hook triggers a similar server-side flow: ImageTextExtraction::generate_ocr_text() calls run(), which engages the OpenAI provider to get the text, and then save() stores it in the attachment's post_content.