Task:
    Create a Streamlit application named app.py for a "YouTube Content Reaction Study."

Context:
    - I already have a .env file containing my OPENAI_API_KEY
    - The app must load video data from a youtube_data.json file
    - The app must send captured images to the gpt-5-nano model using the OpenAI SDK

Technical Requirements:

1. Cross-Platform Webcam Capture:
   - Use OpenCV (cv2) for capturing images
   - Important: Ensure the code is cross-platform
     * Use a check like: if sys.platform == "win32": cap = cv2.VideoCapture(0, cv2.CAP_DSHOW)
     * Otherwise use: cv2.VideoCapture(0)
   - Capture images every 10 seconds while the "Start Recording" toggle is on
   - Limit to a maximum of 20 images

2. Streamlit UI:
   - Display the YouTube video using the iframe HTML from the JSON
   - Create a 2-column layout:
     * Column 1: "Start Recording" toggle and an "Evaluate Response" button
     * Column 2: Show the current image capture count

3. Image Handling:
   - Save images to an /images folder
   - Automatically clear this folder whenever a new video is selected from the dropdown

4. AI Logic:
   - Format a prompt using a template from prompt_reaction.txt
   - Send the formatted text and the base64-encoded images to OpenAI

5. Logging:
   - Log all major events to the server terminal with timestamps:
     * Camera starts
     * Images saved
     * AI calls