Batch-extract Diablo II: Resurrected item tooltips from screenshots using a Vision LLM via an OpenAI-compatible API (e.g., Groq, OpenAI, OpenRouter, LM Studio/Ollama OpenAI bridge).

• Batch Processing: Process hundreds of item screenshots automatically
• Smart OCR: Uses AI vision models to accurately read item tooltips
• Clean Output: Removes UI elements like “Ctrl + Left Click to Move”, set lists, and similar boilerplate
• Item Categorization: Automatically categorizes items (WEAPON, ARMOR, RING, etc.)
• Searchable Database: Browse and filter your entire item collection
• Virtualized Catalog: Smooth, infinite-like scrolling for large lists (350+ items)
• Debounced Search: Filters update after you pause typing for a smoother feel
• Persistent Settings: Your Item Tracker settings are auto-saved and restored
• Parallel Workers: Tune Workers to control how many requests run in parallel

1. Point to Screenshots: Select your D2R screenshots folder
2. Configure API: Set up your vision AI service (Groq recommended - it's free!). After you set Endpoint, Model, and API Key, configure:

• Workers – How many screenshots to process in parallel (1–10). Higher = faster but more likely to hit provider rate limits.
• Rate Limit RPM & Jitter – Keep within your provider’s limits.

3. Batch Process: The AI reads each screenshot and extracts the item text
4. Clean Results: Automatically removes UI clutter and formats the output
5. Save to File: All items saved to a single text file, separated by ---

1. Load Items: Import your processed item text files
2. Smart Search: Type to search; results update after ~0.3s pause
3. Category Filter: Filter by item type (weapons, armor, charms, etc.)
4. Virtualized Scrolling: The list renders only visible items, so it stays smooth even with very large collections

See the examples/ folder for sample screenshots and their expected output format.

1. Get the EXE

   • Open the repository's Releases page on GitHub
   • Download D2R-AI-Item-Tracker.exe from the latest release (Assets section)

2. Run the App

   • Double-click D2R-AI-Item-Tracker.exe
   • In the Item Tracker tab, fill in:
     • Endpoint – OpenAI-compatible chat-completions URL
       (e.g., https://api.groq.com/openai/v1/chat/completions)
     • Model – vision-capable model id
       (e.g., meta-llama/llama-4-scout-17b-16e-instruct)
     • API Key – your API key
     • Screenshots folder – your D2R screenshots directory
     • Output file – where to save results (defaults to output.txt)
   • Click Run. The log shows progress as each screenshot is processed

3. Browse Items

   • Switch to the Item List tab
   • Choose a folder containing your .txt exports (where output.txt is saved)
   • Click Load Items
   • Use the Search box and Item Type dropdown to filter your collection

Groq provides free access to hosted LLaMA vision models without requiring a credit card.

1. Go to https://console.groq.com
2. Sign in with a Google account
3. Open API Keys in the left menu
4. Click Create API Key → Copy your gsk_... token

• Endpoint: https://api.groq.com/openai/v1/chat/completions
• Model: meta-llama/llama-4-scout-17b-16e-instruct
• API Key: Your gsk_... key

That's it! No billing setup required. If you hit rate limits, just wait a bit and retry.

The AI automatically categorizes items into these types:

• "fire resist" → Find all items with fire resistance
• "shako" → Find Harlequin Crest
• "+skills" → Find items with +skill modifiers
• Filter by WEAPON + search "bow" → Find all bows

📁 Your D2R Screenshots/
├── Screenshot001.png    ← Item tooltips
├── Screenshot002.png    
└── output.txt          ← Generated results

📁 examples/
├── screenshots/        ← Sample item screenshots  
└── output/Akuma-javazon.txt ← What the AI should produce

Output Format (output.txt):

PLAGUE FINGER
CHAIN GLOVES
DEFENSE: 16
DURABILITY: 16 OF 16
REQUIRED STRENGTH: 25
REQUIRED LEVEL: 72
+2 TO JAVELIN AND SPEAR SKILLS (AMAZON ONLY)
+20% INCREASED ATTACK SPEED
+63% ENHANCED DEFENSE
HALF FREEZE DURATION
LEVEL 1 POISON NOVA (22/22 CHARGES)
[CATEGORY: ARMOR]
---
RUBY GRAND CHARM OF LIFE
REQUIRED LEVEL: 22
+11 TO LIFE
FIRE RESIST +2%
[CATEGORY: CHARM]

Any OpenAI-compatible vision API will work:

Windows Controlled Folder Access may block saving in protected folders (Documents, Desktop).

Solutions:

• Choose a different output folder (e.g., Downloads)
• Allow the app: Windows Security → Ransomware Protection → Allow an app through Controlled folder access
• The app auto-fallbacks to: %LOCALAPPDATA%\D2R-AI-Item-Tracker\

• If you see many 429 errors:
  • Lower Workers and/or Rate Limit (RPM).
  • Keep Jitter > 0 to stagger calls.
  • The app automatically re-queues 429’d screenshots for a later attempt.

Items are automatically cached in %LOCALAPPDATA%\D2R-AI-Item-Tracker\. Use Clear List to reset.

• "No images found" → Check screenshots folder path
• "API key missing" → Verify your API key is correct
• "2 chars output" → AI couldn't read the screenshot, try different images

• Python 3.9+
• pip install: opencv-python numpy requests Pillow python-dotenv

git clone https://github.com/vdamov/D2R-AI-Item-Tracker.git
cd D2R-AI-Item-Tracker
pip install -r requirements.txt
pyinstaller --onefile --hidden-import=tkinter --name "D2R-AI-Item-Tracker" --icon=assets/images/icon.ico --add-data "assets/fonts;assets/fonts" --add-data "assets/images/icon-color.ico;images" main.py

The repository includes automated EXE building via GitHub Actions. Each release automatically generates the Windows executable.

• High resolution: 1080p+ screenshots work best
• Full tooltip visible: Make sure item tooltip is completely shown
• Good contrast: Dark background, bright tooltip text
• No overlapping: Avoid overlapping windows or UI elements

• Organize by character: Use separate folders for each character's items
• Remove duplicates: Delete blurry or duplicate screenshots first
• Use meaningful names: Name output files by character (e.g., javazon_items.txt)

• Use quotes: "faster cast rate" for exact phrases
• Combine filters: Set category to WEAPON then search for specific weapon types
• Character names: Search by character name to find items from specific heroes

MIT License © 2025 Vladimir Damov

Pull requests welcome! Please see the examples/ folder for expected input/output formats when testing changes.