datpat/jellyfin-orion-o6

This project packages a customized version of Jellyfin, a self hosted alternative to services like Netflix for streaming your own video library, so that it can use the dedicated video processing chip built into a specific single board computer, the Radxa Orion O6. Normal video streaming software often has to convert video files on the fly using the computer's main processor, which can get slow or run hot on smaller devices. This project instead teaches Jellyfin to hand that conversion work off to a piece of hardware built specifically for it, called a VPU, which is much more efficient. To make that work, the project combines three pieces. First, it builds a customized version of FFmpeg, the underlying video conversion engine that Jellyfin relies on, using patches written for this particular chip so it can talk to the hardware video decoder and encoder. Second, when that hardware path is not available or not chosen, it also builds its own optimized versions of several common software video encoders, tuned specifically for this chip's processor cores, so quality stays high even without the hardware shortcut. Third, and most involved, it applies patches directly to Jellyfin's own server and web interface, because the stock version of Jellyfin only knows how to use this kind of hardware acceleration for one video format. The patches teach it to use the hardware for several more formats, add a toggle in the settings to turn the newer hardware encoding on, and fix the on screen playback statistics so they correctly report whether hardware or software is actually doing the work. Everything ships as a Docker image, a self contained package that bundles the software with everything it needs to run, which is built using a provided script and then deployed onto the Orion O6 itself with a docker compose file that grants it access to the video hardware. Once running, the hardware options appear in Jellyfin's normal transcoding settings page, and a log message during playback confirms whether the hardware or software path was used. The project notes some rough edges, such as the hardware encoder producing slightly lower quality per file size than the tuned software encoder, though it processes video much faster and uses far less of the main processor. Jellyfin itself is licensed under the GPLv3 license, and this project's patches build directly on top of that codebase.