ffmpeg screen recording compression: downscaling and CRF cut size, a slower preset or halving fps barely does
Measured on 20 s synthetic 2560x1440 60 fps clips (lavfi testsrc2 with full-frame motion, 185.6 MB source, and a mostly static frame with a small moving window, 22.8 MB source), so treat the ratios as indicative rather than exact. libx264 -crf 23 at native size gave 34.4 MB / 3.2 MB. Adding -vf scale=-2:1080 gave 11.1 / 2.2 MB (-68% / -32%); adding fps=30 on top gave 9.1 / 1.7 MB, only about 20% more. CRF 28 instead of 23 at native size: 16.4 / 2.2 MB; CRF 18: 62.5 / 5.1 MB. -preset slow did not shrink anything at the same CRF (34.9 vs 34.4 MB, 16 s vs 12 s to encode), while -preset veryfast was smaller and twice as fast (29.5 MB, 5.7 s) at slightly lower SSIM (0.990 vs 0.992). libx265 -crf 28 took 49 s where libx264 took 12 s. The command that took 185.6 MB to 9.1 MB: ffmpeg -i in.mov -vf 'scale=-2:1080,fps=30' -c:v libx264 -crf 23 -pix_fmt yuv420p -c:a aac -b:a 128k -movflags +faststart out.mp4. Tested with ffmpeg 6.0 (npm ffmpeg-static 5.3.0) on Apple Silicon, 2026-11-15.