I used my local LLM to sort hundreds of gaming clips, and it was the laziest solution that worked

A developer wrote a Python script using Gemma 4 31b running locally via llama.cpp to automatically sort hundreds of gaming clips into organized folders. The script extracts three frames from each video, sends them as base64-encoded images to an OpenAI-compatible API, and uses a majority-vote system to identify the game. The approach proved reliable for distinguishing visually distinct games like Counter-Strike 2 and Valorant, with dry-run and month-filter flags added for safety and convenience.