slevin7 86 Posted December 28, 2025 Posted December 28, 2025 Currently, the "Video preview thumbnail extraction" task uses software decoding via ffmpeg, which is very CPU-intensive and slow. On my server with a large library (~30TB), the task estimates approximately 80+ hours to complete, even though I have Intel QuickSync (QSV) available that Emby already uses successfully for transcoding. Current behavior: /bin/ffmpeg -f matroska -threads 1 -skip_interval 10 -i file:... -an -sn -vf scale=w=320:h=180 -f image2 ... The extraction runs single-threaded with software decoding only. Proposed enhancement: Add an option to use hardware-accelerated decoding for chapter/preview image extraction, similar to how transcoding already supports VAAPI, QSV, and NVENC. Example with VAAPI/QSV: /bin/ffmpeg -hwaccel vaapi -hwaccel_output_format vaapi -i file:... -vf 'scale_vaapi=w=320:h=180,hwdownload,format=nv12' -f image2 ... Benefits: - Significantly faster extraction (estimated 5-10x speedup) - Lower CPU usage during extraction - Better utilization of existing hardware that's already configured for transcoding - Users with large libraries could complete the task in hours instead of days Since Emby already detects and configures hardware acceleration for transcoding, it would be great if the same hardware could optionally be used for thumbnail extraction. Thank you for considering this enhancement! 1
Happy2Play 9780 Posted December 28, 2025 Posted December 28, 2025 We had a discussion over here awhile ago as to there really being no benefit (at least at that time).
Recommended Posts