V 3.1

🚀 OCRAIHub v3.1 — The Multi-Engine AI Document Parsing Release

🌟 What's New

• Baidu PaddleOCR-VL (v1.6): Fully upgraded to the official Asynchronous Job API (/api/v2/ocr/jobs) with support for specialized task types (ocr, table, formula, chart) and automated retry handling during peak queue traffic (Code 10010).
• TextIn (v1 xparse): Integrated TextIn’s latest document parsing engine with automated DPI validation (72, 144, 216) and structured Markdown table extraction.
• MinerU (vlm): Added batch processing for MinerU, featuring automated presigned upload URLs, asynchronous polling, and automated ZIP artifact extraction.

📄 Advanced Document & PDF Features

• 📄 Native PDF Mode: Send raw .pdf files directly to Baidu, TextIn, or MinerU for server-side native parsing, bypassing local rendering.
• 🖼️ Auto-Download Figures: Automatically detects, crops, and downloads inline figures, charts, and diagrams into a dedicated local folder ([filename]_[model]_images).
• ✂️ Auto-Column Slicing & Deskewing: Pre-processes scanned book pages and multi-column documents (dictionaries, academic papers) with projection profiling and Otsu binarization.
• 🛑 Instant Task Cancellation: Added a responsive 🛑 Stop button that immediately halts background threads and long-running API polling without locking the UI.

💾 Smarter Auto-Save & File Management

• Dynamic Output Naming: Markdown outputs and extracted image folders are now automatically tagged with the active engine name (e.g., Document_gemini-2.5-flash.md, Document_MinerU.md, Document_Textin.md).
• Batch Image Auto-Save: Processing batches of standalone images now automatically aggregates and saves results directly to Markdown.

---

🛠️ Bug Fixes & Stability

• Fixed NoneType String Concatenation: Resolved a crash where empty or safety-blocked model responses caused a TypeError during document assembly.
• Resolved MAX_TOKENS Failure: Models hitting output token ceilings now preserve and output all generated text up to the limit rather than discarding it.
• PyMuPDF Modernization: Updated deprecated fitz namespace calls to import pymupdf to eliminate terminal warnings.
• macOS Font Optimization: Added native monospace font fallbacks (Menlo, Monaco) to remove startup font scanning delays on macOS.

---

🎨 UI / UX Upgrades

• Single-Line File Inputs: Replaced bulky preview boxes with compact, native file path inputs.
• Smart Engine Swapping: Dynamically displays only the required credential and configuration fields for the active engine.
• Collapsible Panels: Added toggle switches for Advanced Pre-processing Settings and Prompt Instructions.
• Gravity-Anchored Layout: Applied layout anchors to prevent gaps or visual jumping across macOS, Windows, and Linux.

---

📜 License & Citation

This project is licensed under the GNU General Public License v3.0 (GPLv3). If you use this software in your research or workflows, please refer to the root CITATION.cff file.

V3.0

🚀 Release Notes: v3.0 - The Multi-Engine Update

This major update introduces powerful new AI models, native PDF handling, and a highly polished UI designed for professional document digitization.

🌟 New AI Engines Supported

• Gemini 3.5 flash (gemini-flash-latest), gemini-3.1-pro-preview and gemini-3.1-flash-lite, gemini-3-flash-preview
• implemented 🛑 Stop button
• Baidu PaddleOCR-VL (v1.6): Integrated full support for Baidu's industry-leading layout parsing model.
  • Asynchronous API: Fully migrated to Baidu's Async API (/api/v2/ocr/jobs) to prevent 500 Server Errors when processing large documents.
  • Baidu Tasks: Added a dropdown to select specific tasks: ocr, table, formula, or chart.
• MinerU: Integrated the highly accurate MinerU Cloud API (vlm model) for complex document parsing.
  • Batch URL Polling: Fully implemented MinerU's strict batch processing flow with automated upload URLs and polling.
• Textin: Added support for the Textin API (pdf_to_markdown), enabling lightning-fast extraction of document trees and tables.
  • Smart DPI Snapper: Added automatic DPI validation (snaps to 72, 144, or 216) to comply with strict Textin API requirements.

📄 Advanced Document Handling

• Native PDF Mode: Added a toggle to send raw .pdf files directly to Baidu, Textin, or MinerU. This bypasses local PyMuPDF rendering, allowing the remote AI engines to handle the native file formats directly for superior results.
• Auto-Download Figures: (Baidu & MinerU) The app now automatically reads the JSON response, detects extracted charts/figures/inline images, and downloads them directly into a local folder next to your output file.

💾 Smarter Auto-Save & Export

• Dynamic Auto-Save Naming: Auto-saved .md files and image folders are now automatically tagged with the exact engine and model used (e.g., Document_gemini-2.5-flash.md or Document_MinerU.md). This makes comparing different models incredibly easy.
• Image Batch Auto-Save: The auto-save feature now works for standalone image batches as well as PDFs.

🖥️ UI / UX Overhaul

• Compact Single-Line Inputs: Replaced bulky, empty grey boxes with sleek, single-line text inputs for file selection, saving massive amounts of vertical space.
• Smart Engine Swapping: Selecting an engine from the dropdown now instantly updates the required input fields (e.g., swapping "Custom URL" for "Secret Code" or updating the Proxy box), preventing user confusion.
• Collapsible Settings: Added toggle buttons (⚙️ Show Advanced Settings and 💬 Show Prompt Settings) to keep the interface clean while retaining all advanced options.
• The "Gravity Anchor": Implemented a dynamic UI stretch factor that ensures all buttons stay tightly grouped at the top of the screen when menus are collapsed, eliminating ugly grey gaps.
• Robust Connection Testing: Updated the Test Connection button to send a tiny, dynamically drawn 1x1 pixel image/payload to ensure accurate connectivity testing for PaddleOCR, Textin, and MinerU without causing 500 errors.

Full Changelog: v2.0...V3.0

OCR AI Hub V2

Screenshots

---

For MacOS user:

Solution 1: The "Right-Click" Method (Easiest & Most Common):

1. Move the app to the Applications folder.

2. Do not double-click to open it the first time.

3. Instead, Right-Click (or Control-Click) on the app icon.

4. Select Open from the menu.

5. A warning window will appear; click the Open button again. (This step is only required once. Afterwards, the app will open normally with a double-click).

---

Solution 2: The Terminal Fix (For Advanced Users):

xattr -cr /Applications/OCRAIHub.app