FlixBus CitySnap: How we use GenAI and not only to collect captivating images for cities and confir…

Discover how FlixBus CitySnap uses GenAI to automatically collect, validate, and manage city photos across 5000+ locations, combining LLMs, landmark detection, and smart cropping.

Key takeaways
  • FlixBus CitySnap project aims to automatically collect and validate attractive city photos across 5,000+ cities where Flix operates

  • The system uses a combination of:

    • ChatGPT/Gemini to identify city attractions
    • Image stock APIs (Pixabay, Pixels, Wikimedia) with CC0 licensed images
    • Google Landmark Detection and Gemini for photo validation
    • Smart cropping tools for image resizing
  • Key technical insights:

    • Gemini outperforms ChatGPT for landmark detection in small/medium cities
    • Google Landmark Detection works better for major cities
    • Using two different prompts improves accuracy: one for attraction identification, one for validation
    • Smart cropping preserves image quality better than center cropping
  • Business applications:

    • Marketing/social media campaigns
    • Internal photo library for stakeholders
    • Promotional materials
    • E-commerce platform enhancement
  • Important considerations:

    • Trade-off between precision and recall when collecting images
    • Need to filter out transport vehicles from photos
    • Must verify photo locations to avoid misclassification
    • Automated pipeline crucial for scalability
    • Regular model updates as new LLM versions release
  • Images are stored in four standardized sizes to maintain consistency across platforms

  • Using only CC0 licensed images eliminates attribution requirements and legal issues