Skip to content

PaddleOCR-VL 1.6 Setup

This guide shows how to run PaddleOCR-VL 1.6 locally or through the Baidu API in Zotero Research Assistant.

Local mode keeps PDF data on your machine. API mode sends PDF document data to Baidu's servers for processing.

1) Choose local or API mode

Local worker mode

  1. Keep the default Python worker enabled under Prerequisites.
  2. Start or recreate the Redis stack so the worker installs the updated dependencies.
  3. Under OCR, select PaddleOCR-VL 1.6 (local).

The worker pins paddleocr[doc-parser] 3.6.0, PaddleX 3.6.1, and PaddlePaddle 3.2.2. The 1.6 model files download on first use and are stored in the configured worker cache.

For the advanced legacy local runtime, update the environment first:

source .venv/bin/activate
pip install -r requirements.txt

API mode

  1. Create or sign in to your PaddleOCR account at AI Studio PaddleOCR.
  2. Select PaddleOCR-VL 1.6 and copy its API token and API URL.
  3. Keep that token and URL ready for plugin settings.

2) Configure API settings in the plugin

In Obsidian plugin settings:

  1. Open OCR.
  2. Set Paddle OCR API key to your token.
  3. Set PaddleOCR-VL API URL to your URL.
  4. Set OCR engine to PaddleOCR-VL 1.6 (API).

Notes:

  • The engine option appears after a non-empty API key is saved.
  • Keep PP-StructureV3 API URL empty unless you also plan to use PP-Structure API mode.

For mixed PDF quality:

  • OCR decision (when to OCR): OCR only if text is poor
  • OCR layout override (per-page): keep off unless needed for difficult layouts

For mostly scanned PDFs:

  • OCR decision (when to OCR): Prefer OCR for full document

4) Quick verification

  1. Import one short PDF.
  2. Open logs if needed (Maintenance → Logs).
  3. Confirm OCR runs with PaddleOCR-VL 1.6 and no model, token, or HTTP errors.

Common pitfalls

  • Engine option missing: API key is empty or not saved.
  • Local model option missing: update the local Python environment from requirements.txt, or use the default worker mode.
  • Long first local run: the worker must download the PaddleOCR-VL 1.6 model files before inference.
  • 401/403 auth error: invalid or expired API token.
  • 429 responses: API quota/rate limit reached.
  • Wrong endpoint: use the .../ocr/layout-parsing URL for this plugin flow.
  • Network/proxy issues: Obsidian cannot reach aip.baidubce.com.