Developer creates virtual cycling software using video AI

A developer has designed a custom virtual cycling application that uses video-generating AI to create highly realistic environments for indoor training. Unlike traditional 3D-modeled platforms, this system utilizes video models such as MiniMax H3 and LTX 2.5 to generate lifelike footage of real-world locations, such as Broadway or Shakujii Park.

The technical approach involves decoupling video generation from playback to ensure smooth performance. The system pre-generates video segments based on OpenStreetMap data and real-world photos from Mapillary, then adjusts the playback speed to match the user's physical input, such as cadence and power, received via Bluetooth FTMS from a smart trainer. This method allows for a more immersive experience by simulating real-world terrain and gradients through physical calculations.