OCR & Text Extraction

Bayanat extracts text from images attached to Bulletins. Extracted text is stored alongside the media and becomes searchable.

Supported Formats

PNG, JPEG, TIFF, JPG, GIF, WebP, BMP, PNM.

How It Works

1. Upload image files to a Bulletin as media attachments
2. Trigger OCR from the Bulletin's media panel (individual or all pending)
3. Extraction runs asynchronously in the background
4. Results appear on each media item with: extracted text, confidence score, word count, and detected language

Default language hints are Arabic and English (configurable).

OCR Providers

Bayanat supports two OCR providers, configured via the OCR_PROVIDER environment variable.

Google Vision (google_vision)

Google's cloud OCR service. Returns confidence scores, detected language, orientation correction, and bounding box data used by the Text Map feature.

OCR_PROVIDER=google_vision
GOOGLE_VISION_API_KEY=your-api-key

LLM (llm)

Works with any OpenAI-compatible /v1/chat/completions endpoint. This includes self-hosted models (Ollama, vLLM, SGLang) and cloud APIs (OpenAI, OpenRouter).

OCR_PROVIDER=llm
LLM_OCR_URL=http://localhost:11434
LLM_OCR_MODEL=llava
LLM_OCR_API_KEY=               # optional, for authenticated endpoints

Example configurations:

Provider	URL	Model
Ollama (local)	http://localhost:11434	llava
vLLM on RunPod	https://{pod_id}-8000.proxy.runpod.net	Qwen/Qwen2.5-VL-72B-Instruct-AWQ
OpenAI	https://api.openai.com	gpt-4o
OpenRouter	https://openrouter.ai/api	anthropic/claude-sonnet-4-20250514

The LLM provider includes automatic EXIF orientation correction and image downscaling for large files.

TIP

The Text Map visual overlay feature is only available with the Google Vision provider, as it requires bounding box coordinates that LLM endpoints don't return.

Bulk Processing

OCR can be triggered in bulk for multiple Bulletins at once, processed asynchronously via the task queue. The system rate-limits requests to stay within API quotas (1200 requests/minute).

Review and Editing

After extraction, users can:

• View the extracted text and confidence score
• Edit the extracted text manually (original is preserved)
• View a Text Map overlay showing text locations on the image (Google Vision only)
• Search across all extracted text

Configuration

See Configuration for full setup details.