🔍 DeepSeek-OCR-WebUI

Visit Application →

🎉 v4.1 Update: UI Improvements & Model Version Display

• 🏷️ OCR-2 Model Badge — Header now shows a prominent OCR-2 badge so users instantly know the model version
• 🎨 Table Rendering Fix — OCR-detected tables now display with white backgrounds, dark text, and zebra striping for clear readability (previously appeared as dark/unreadable blocks)
• 📡 Health API model_version — /health endpoint now returns "model_version": "DeepSeek-OCR-2" for programmatic version detection
• 🔖 Footer Version — Updated to v4.1 · OCR-2

🎉 v4.0 Update: DeepSeek-OCR-2 Model Upgrade!

🚀 Major model upgrade to DeepSeek-OCR-2 (Visual Causal Flow) — better accuracy, higher resolution!

✨ What's New in v4.0

• 🧠 DeepSeek-OCR-2 Model - Upgraded to the latest DeepSeek-OCR-2 with Visual Causal Flow architecture
• 🔬 Higher Resolution - Dynamic resolution up to (0-6)×768×768 + 1×1024×1024 (was 640×640)
• ⚡ Flash Attention 2 - Native flash_attention_2 support on CUDA for optimal inference speed
• 🎯 Improved Accuracy - Better document understanding, chart parsing, and text recognition
• 🔄 Full Backward Compatibility - All 7 recognition modes, REST API, and frontend unchanged
• 🐳 Docker v4.0 - New all-in-one image with pre-downloaded OCR-2 model (Dockerfile.v4.0)
• 📦 Unified Tokenizer - Switched from AutoProcessor to AutoTokenizer (aligned with official OCR-2 API)

🔧 Technical Changes

💡 All existing features from v3.6 (concurrency, rate limiting, queue management, Vue 3 frontend) are fully preserved.

🎉 v3.6 Update: Backend Concurrency & Rate Limiting!

🚀 Performance optimization with smart queue management and rate limiting!

✨ What's New in v3.6

• ⚡ Backend Concurrency Optimization - Non-blocking inference with ThreadPoolExecutor
• 🔒 Rate Limiting - Per-client and per-IP request limits (X-Client-ID header support)
• 📊 Queue Management - Real-time queue status with position tracking
• 🏥 Enhanced Health API - Queue depth, status (healthy/busy/full), and rate limit info
• 🌐 New Languages - Added Traditional Chinese (zh-TW) and Japanese (ja-JP)
• 🎯 429 Error Handling - Graceful handling when queue is full or rate limited

🙏 Contributors: @cloudman6 (PR #41)

🎉 v3.5 Major Update: Brand New Vue 3 Frontend!

🚀 Complete UI Overhaul with Modern Vue 3 + TypeScript Architecture!

Home Page	Processing Page

✨ What's New in v3.5

• 🎨 Brand New Vue 3 UI - Modern, responsive design with Naive UI components
• ⚡ TypeScript Support - Full type safety and better developer experience
• 📦 Dexie.js Database - Local IndexedDB for offline page management
• 🔄 Real-time Processing Queue - Visual OCR progress with queue management
• 🏥 Health Check System - Backend status monitoring with visual indicators
• 📄 Enhanced PDF Support - Smooth PDF rendering with page-by-page processing
• 🌐 i18n Ready - Built-in internationalization (EN/CN/TW/JP)
• 🧪 E2E Testing - Comprehensive Playwright test coverage

👥 Contributors

This project is the result of an outstanding collaboration. The Vue 3 frontend was developed through a successful merge of PR #34.

CloudMan 🏆 Vue 3 Frontend Lead Developer 164 commits · Complete UI Rewrite

neosun100 🎯 Project Maintainer Backend · Docker · Integration

💡 About the Vue 3 Frontend: @cloudman6 contributed an exceptional Vue 3 + TypeScript frontend with 164 commits, including comprehensive E2E tests, modern UI components, and production-ready architecture. This collaboration transformed DeepSeek-OCR-WebUI into a professional-grade application!

📖 Introduction

DeepSeek-OCR-WebUI is an intelligent document recognition web application powered by the DeepSeek-OCR model. It provides a modern, intuitive interface for converting images and PDFs to structured text with high accuracy.

✨ Core Highlights

🚀 Features

7 Recognition Modes

🆕 Vue 3 Frontend Features

┌─────────────────────────────────────────────────────────────┐
│  📁 Page Sidebar          │  📄 Document Viewer             │
│  ├─ Thumbnail List        │  ├─ High-res Image Display      │
│  ├─ Drag & Drop Reorder   │  ├─ OCR Overlay Toggle          │
│  ├─ Batch Selection       │  ├─ Zoom Controls               │
│  └─ Quick Actions         │  └─ Status Indicators           │
├─────────────────────────────────────────────────────────────┤
│  🔄 Processing Queue      │  📝 Result Panel                │
│  ├─ Real-time Progress    │  ├─ Markdown Preview            │
│  ├─ Cancel/Retry          │  ├─ Word/PDF Export             │
│  └─ Health Monitoring     │  └─ Copy to Clipboard           │
└─────────────────────────────────────────────────────────────┘

🖼️ Screenshots

Home Page

Processing Interface

Quick Start Guide

📦 Quick Start

🐳 Docker (Recommended)

# Pull and run
docker pull neosun/deepseek-ocr:v4.1
docker run -d \
  --name deepseek-ocr \
  --gpus all \
  -p 8001:8001 \
  --shm-size=8g \
  neosun/deepseek-ocr:v4.1

# Access: http://localhost:8001

Available Docker Tags

🍎 Mac (Apple Silicon)

# Clone and setup
git clone https://github.com/neosun100/DeepSeek-OCR-WebUI.git
cd DeepSeek-OCR-WebUI

# Create conda environment
conda create -n deepseek-ocr python=3.11
conda activate deepseek-ocr

# Install dependencies
pip install -r requirements-mac.txt

# Start service
./start.sh
# Access: http://localhost:8001

🐧 Linux (Native)

# With NVIDIA GPU
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt
./start.sh

🔌 API & Integration

REST API

import requests

# Single image OCR
with open("image.png", "rb") as f:
    response = requests.post(
        "http://localhost:8001/ocr",
        files={"file": f},
        data={"prompt_type": "ocr"}
    )
    print(response.json()["text"])

# PDF OCR (all pages)
with open("document.pdf", "rb") as f:
    response = requests.post(
        "http://localhost:8001/ocr-pdf",
        files={"file": f},
        data={"prompt_type": "document"}
    )
    print(response.json()["merged_text"])

Endpoints:

• GET /health - Health check
• POST /ocr - Single image OCR
• POST /ocr-pdf - PDF OCR (all pages)
• POST /pdf-to-images - Convert PDF to images

📖 Full API Documentation: API.md

MCP (Model Context Protocol)

Enable AI assistants like Claude Desktop to use OCR:

{
  "mcpServers": {
    "deepseek-ocr": {
      "command": "python",
      "args": ["/path/to/mcp_server.py"]
    }
  }
}

📖 MCP Setup Guide: MCP_SETUP.md

🌐 Multilingual Support

Switch language via the selector in the top-right corner.

📊 Version History

v4.1 (2026-02-20) - UI Improvements & Model Version Display

🏷️ UI & API Enhancements:

• ✅ OCR-2 model badge in header for instant version recognition
• ✅ Table rendering fix: white background, dark text, zebra striping
• ✅ Health API returns model_version: "DeepSeek-OCR-2"
• ✅ Footer updated to v4.1 · OCR-2

v4.0 (2026-02-20) - DeepSeek-OCR-2 Model Upgrade

🧠 Major Model Upgrade:

• ✅ Upgraded to DeepSeek-OCR-2 (Visual Causal Flow)
• ✅ Dynamic resolution: (0-6)×768×768 + 1×1024×1024
• ✅ Flash Attention 2 on CUDA for optimal inference speed
• ✅ Switched from AutoProcessor to AutoTokenizer
• ✅ image_size upgraded from 640 to 768
• ✅ New Dockerfile.v4.0 with pre-downloaded OCR-2 model
• ✅ Full backward compatibility with all v3.6 features

v3.6 (2026-01-20) - Backend Concurrency & Rate Limiting

⚡ Performance Optimization:

• ✅ Non-blocking inference with ThreadPoolExecutor
• ✅ Concurrency control with asyncio.Semaphore (OCR: 1, PDF: 2)
• ✅ Queue system with MAX_OCR_QUEUE_SIZE and dynamic status
• ✅ Per-IP and per-Client-ID rate limiting (X-Client-ID header)
• ✅ 429 error handling (queue full, client limit, IP limit)
• ✅ Health indicator with 3 status colors (green/yellow/red)
• ✅ OCR queue popover with real-time position display

🙏 Contributors: @cloudman6 (PR #41)

v3.5 (2026-01-17) - Vue 3 Frontend

🎨 Complete UI Overhaul:

• ✅ Vue 3 + TypeScript + Naive UI
• ✅ Dexie.js local database
• ✅ Real-time processing queue
• ✅ Health check monitoring
• ✅ E2E test coverage (Playwright)
• ✅ GitHub links in header

🙏 Contributors: @cloudman6 (164 commits)

v3.3.1 (2025-12-16) - BFloat16 Fix

• ✅ Fixed GPU compatibility for RTX 20xx, GTX 10xx
• ✅ Auto-detect compute capability

v3.3 (2025-11-05) - Apple Silicon

• ✅ Native MPS backend for Mac M1/M2/M3/M4
• ✅ Multi-platform architecture

v3.2 (2025-11-04) - PDF Support

• ✅ PDF upload and conversion
• ✅ ModelScope auto-fallback

📖 Documentation

📈 Star History

🤝 Contributing

Contributions welcome! Please:

1. Fork this repository
2. Create feature branch (git checkout -b feature/AmazingFeature)
3. Commit changes (git commit -m 'Add AmazingFeature')
4. Push to branch (git push origin feature/AmazingFeature)
5. Open Pull Request

📄 License

This project is licensed under the MIT License.

🙏 Acknowledgments

• DeepSeek-AI - DeepSeek-OCR model
• @cloudman6 - Vue 3 frontend development
• All contributors and users