Developer builds zero-GPU ASMR rain video pipeline using Python and ffmpeg

A developer has created a fully automated system that generates 30-minute ASMR rain videos daily at virtually no cost, using only a still image, a Python-based raindrop physics simulation, and ffmpeg's displace filter. The pipeline eliminates the need for AI video generation tools like Sora or Runway, which would cost hundreds to thousands of yen per video. A key technical challenge was creating seamless 16-second loop cycles that tile into 30-minute videos without visible seams, achieved by ensuring displacement maps match exactly at the start and end frames. The system covers seven interior themes — such as a café window seat and a Japanese room — with automated masking that detects window and fire regions to apply effects accurately. The developer reports the daily build job now completes in just over four minutes, down from seven to eight minutes, and the setup runs autonomously via a scheduled launchd job.