FFmpeg video compression benchmark

There are many modern video encoding algorithm that can be used for video compression. This article tests several of them and gives a simple benchmark.

Benchmark setup

The video to compress is the lossless version of the trailer of the short film Sintel. The video file can be found here. I use the 1080p version of the video.

The video will be encoded with H.264, H.265, VP9 and AV1 encoding algorithms on difference options.

The results will contain the size of the encoded video, the compression time, three screen shorts at the same frame and a subjective view for the video. The encoding is performed on a laptop with other running programs so the compression times will be less representative. The screenshots are taken at the 13th second, 17th second and 39th second with the following bash script:

#!/usr/bin/env bash
if [ $# -lt 1 ]; then
    echo "Usage: $0 <input_video>"
    exit 1
fi

input_video=$1

ffmpeg -ss 00:00:13 -i "$input_video" -vframes 1 -q:v 2 s13.jpg
ffmpeg -ss 00:00:17 -i "$input_video" -vframes 1 -q:v 2 s17.jpg
ffmpeg -ss 00:00:39 -i "$input_video" -vframes 1 -q:v 2 s39.jpg

Original video

Result

• Size: 3,805,996KB or 3.6GB.
• Compression time: 0s.
• Screenshots:
  • 13th second:
  • 17th second:
  • 39th second:
• Subjective view: Scene transitions are crisp, with rich details and saturated colors.

H.264 with low quality

Command

ffmpeg -i sintel_trailer_2k_1080p24.y4m -c:v libx264 -preset slow -crf 30 -b:v 0 -c:a libopus output.mkv

Result

• Size: 4,840KB or 4.7MB.
• Compression time: 12.95s.
• Screenshots:
  • 13th second:
  • 17th second:
  • 39th second:
• Subjective view: There are noticeable blurry areas near the edges, and the overall scene lacks sharpness. The details of the falling snow are indistinct and difficult to recognize. While the colors are acceptable, they appear slightly dim. Additionally, the action sequence is too fast-paced, making it hard to follow.

H.264 with high quality

Command

ffmpeg -i sintel_trailer_2k_1080p24.y4m -c:v libx264 -preset slow -crf 18 -b:v 0 -c:a libopus output.mkv

Result

• Size: 22,644KB or 22MB.
• Compression time: 29.03s.
• Screenshots:
  • 13th second:
  • 17th second:
  • 39th second:
• Subjective view: The blur on the edges are gone. The sharpness is OK and the details of the falling snow are back. The overall impression is actually quite good.

H.265 with low quality

Command

ffmpeg -i sintel_trailer_2k_1080p24.y4m -c:v libx265 -preset slow -crf 30 -b:v 0 -c:a libopus output.mkv

Result

• Size: 3,436KB or 3.4MB.
• Compression time: 57.96s.
• Screenshots:
  • 13th second:
  • 17th second:
  • 39th second:
• Subjective view: Almost as blurry as the H.264 encoding with high compression rate. With more clear action scenes and a little richer details.

H.265 with high quality

Command

ffmpeg -i sintel_trailer_2k_1080p24.y4m -pix_fmt yuv420p10le -c:v libx265 -preset slow -crf 21 -b:v 0 -c:a libopus -x265-params profile=main10 output.mkv

Result

• Size: 11,200KB or 11MB.
• Compression time: 180.05s.
• Screenshots:
  • 13th second:
  • 17th second:
  • 39th second:
• Subjective view: Very close to the original video. However the dark scenes feel a little dark than the original. I'm not sure about the reason.

AV1 with low quality

Command

ffmpeg -i sintel_trailer_2k_1080p24.y4m -c:v libsvtav1 -crf 26 -preset 2 -quality quality output.mp4

Result

• Size: 7,292KB or 7.1MB.
• Compression time: 364.20s
• Screenshots:
  • 13th second:
  • 17th second:
  • 39th second:
• Subjective view: It has surprisingly good quality considering its size. Only get blurry when there are lots of moving particles. But the falling snow is still quite visible.

AV1 with high quality

Command

ffmpeg -i sintel_trailer_2k_1080p24.y4m -c:v libsvtav1 -crf 21 -preset 2 -quality quality output.mp4

Result

• Size: 9,640KB or 9.4MB.
• Compression time: 384.28s.
• Screenshots:
  • 13th second:
  • 17th second:
  • 39th second:
• Subjective view: Almost as good as the original video. Some less visible details get even harder to see. Generally good.

AV1 with high quality and another encoder

libaom-av1 is another AV1 encoder. Libaom is the reference encoder for the AV1 format. libaom-av1 is often slower than libsvtav1.

Command

ffmpeg -i sintel_trailer_2k_1080p24.y4m -c:v  libaom-av1 -b:v 0 -crf 25 -preset 2 -c:a libopus -f matroska output.mkv

Result

• Size: 6,528KB or 6.4MB.
• Compression time: 104.44mins or 6,266s.
• Screenshots:
  • 13th second:
  • 17th second:
  • 39th second:
• Subjective view: Almost as good as the original video. Some less visible details get even harder to see. Generally good.

VP9 with low quality

Command

ffmpeg -i sintel_trailer_2k_1080p24.y4m -c:v libvpx-vp9 -crf 30 -b:v 0 -c:a libopus output.webm

Result

• Size: 9,064KB or 8.9M.
• Compression time: 86.19s
• Screenshots:
  • 13th second:
  • 17th second:
  • 39th second:
• Subjective view: The quality is surprisingly good. There are noticeable blurry areas but not as much as the low quality H.264 one. The action scenes has a little lower quality.

VP9 with high quality

Command

ffmpeg -i sintel_trailer_2k_1080p24.y4m -c:v libvpx-vp9 -crf 18 -b:v 0 -c:a libopus output.webm

Result

• Size: 19,500KB or 19MB.
• Compression time: 83.71s
• Screenshots:
  • 13th second:
  • 17th second:
  • 39th second:
• Subjective view: Almost as good as the original. Some typical web stream artifacts are visible.

Table of some result data

Type	Size	Time
Original	3,805,996KB	-
H.264 with low quality	4,840KB	12.95s
H.264 with high quality	22,644KB	29.03s
H.265 with low quality	3,436KB	57.96s
H.265 with high quality	11,200KB	180.05s
AV1 with low quality	7,292KB	364.20s
AV1 with high quality	9,640KB	384.28s
AV1 with high quality and another encoder	6,528KB	6266s
VP9 with low quality	9,064KB	86.19s
VP9 with high quality	19,500KB	83.71s

References

1. H.264 Video Encoding Guide
2. H.265/HEVC Video Encoding Guide
3. AV1 Video Encoding Guide
4. Stackexchange.com
5. How can I reduce a video's size with ffmpeg?