FFV1 encoding cheatsheet

FFV1 is a video codec developed within FFmpeg. It is lossless, meaning that it compresses video without introducing quantization degradations. Therefore, FFV1 is a good choice for ​archiving and preservation. A ​lossless codec comparison found FFV1 to be the "most balanced", offering "relatively good speed and high compression".

There are two versions of the codec supported by ffmpeg, version 1 and 3.

FFV1 version 1

The following list contains encoding parameters for FFV1 (version 1):

Name	FFmpeg argument	Valid values	Comments
Coder	-coder	0, 1, 2	0=Golomb Rice, 1=Range Coder, 2=Range Coder (with custom state transition table)
Context	-context	0, 1	0=small, 1=large
GOP size	-g	integer >= 1	Reuse context model within a group of frames to improve compression.

FFV1 version 1 is in every way inferior to version 3; only use it for backward compatibility.

For archival use, GOP-size should be "1". Doing so increases error resilience (frames are lost independently) at the cost of compression ratio.

Examples

Copy audio "as-is" and use FFV1.1 as video codec. Audio stream kept as-is.

Parameters are GOP-size=1, coder=1, context=1:

ffmpeg -i <input_video> \
    -acodec  copy       \
    -vcodec  ffv1       \
    -level   1          \
    -coder   1          \
    -context 1          \
    -g       1          \
    <output_video>

FFV1 version 3

The following list contains encoding parameters for FFV1 (version 3).

Name	FFmpeg argument	Valid values	Comments
Coder	-coder	0, 1, 2	0=Golomb-Rice, 1=Range Coder, 2=Range Coder (with custom state transition table)
Context	-context	0, 1	0=small, 1=large
GOP size	-g	integer >= 1	Reuse context model within a group of frames to improve compression.
Version	-level	1, 3	Select which FFV1 version to use.
Threads	-threads	integer >= 1	The number of threads to use while processing. Adjust this to match how many of your available CPU cores you want to use.
Slices	-slices	4, 6, 9, 12, 16, 24, 30	Each frame is split into this number of slices. This affects multithreading performance, as well as filesize: Increasing the number of slices might speed up performance, but also increases the filesize.
Error correction/detection	-slicecrc	0, 1	0=off, 1=on. Enabling this option adds CRC information to each slice. This makes it possible for a decoder to detect errors in the bitstream, rather than blindly decoding a broken slice.
Multi-pass encoding	-pass	1, 2	1=1st pass, 2=2nd pass. FFV1.3 is able to be encoded in multiple passes, to increase compression efficiency. It requires encoding the file twice, though: The 1st pass is to analyze the video source data and logging the results, and the 2nd pass uses this previously gathered information to achieve a higher compression ratio. More information about multi-pass encoding can be read up in the ​Wikipedia article about "Variable bitrate" (VBR).
Multi-pass logfile	-passlogfile	a filename prefix	This is the prefix of the logfile used for storing the information gathered during previous passes in multi-pass encoding mode. Additional info about passlogfile can be found in ​FFmpeg's documentation about video options.

For archival use:

• GOP-size should be "1". Doing so increases error resilience (frames are lost independently) at the cost of compression ratio.
• CRC should be enabled.
• Multi-pass, if temporary disk space allows, may be used to recoup the compression ratio loss from GOP=1.

Examples

• Copy audio "as-is" and use FFV1.3 as video codec.

Parameters are 8 threads, coder=1, context=1, GOP-size=1, 24 slices and slice-CRC on:

ffmpeg -i <input_video>     \
    -acodec   copy          \
    -vcodec   ffv1 -level 3 \
    -threads  8             \
    -coder    1             \
    -context  1             \
    -g        1             \
    -slices  24             \
    -slicecrc 1             \
    <output_video>

• Encode using 2-pass mode, with audio kept as-is:

1st pass:

ffmpeg -i <input_video>     \
    -an                     \
    -vcodec ffv1 -level 3   \
    -pass 1                 \
    -passlogfile my_passlog \
    -f nut /dev/null

2nd pass:

ffmpeg -i <input_video>     \
    -acodec copy            \
    -vcodec ffv1 -level 3   \
    -pass 2                 \
    -passlogfile my_passlog \
    <output_video>

Read from existing file

In order to check which parameters an FFV1 video was encoded with, you can use the following command:

ffprobe -i <input_video> -debug 1

FFV1 version 1

[ffv1 @ 0x385bb80] ver:1 keyframe:1 coder:0 ec:0 slices:1 bps:8

FFV1 version 3

The output for FFV1.3 might look as follows:

[ffv1 @ 0x22444e0] global: ver:3.4, coder:0, colorspace: 0 bpr:10 chroma:1(1:0), alpha:0 slices:3x2 qtabs:2 ec:1 intra:0 CRC:0x613FED4D

That line translates to this:

Name	Label	FFmpeg argument	Value	Comment
ver	Version	-level	major-version:3 minor-version:4	-leveldetermines major-version. Minor-version is encoder-internal and read only.
coder	Coder	-coder	0	0=Golomb Rice, 1=Range Coder, 2=Range Coder (with custom state transition table)
colorspace	Color space		0	0=YCbCr ("YUV"), 1=JPEG2000_RCT ("RGB")
bpr	Bits per sample	-pix_fmt	10 bits	For example with "yuv422p10le"
slices	Slice count	-slices	3x2 = 6	Shows the layout of slices as "width x height"
ec	Error correction	-slicecrc	1	0=OFF, 1=ON
intra	Intra-frame only	-g	0	0=GOP greater than 1, 1=GOP is 1

External links

• ​ffv1_stats, a comparison of file sizes by FFV1 version, slice count, and GOP size.
  • First set compares GOP=1 against GOP=300. Second set compares raw against GOP=1. (The actual column names encode a lot more information: for example, v3c11s30crc1p2 is probably short for -level 3 -coder 1 -context 1 -slices 30 -slicecrc 1, two passes.)
  • The results suggest that intra-prediction works better when the slices is smaller (i.e. higher slice count, smaller input video).