x264

videolan/x264

code.videolan.org

No commits since Sep 2025

x264, the best and fastest H.264 encoder

Last commit

Sep 10, 2025

License

GPL-2.0-or-later

x264 is VideoLAN’s open-source encoder for H.264 / AVC, available as a command-line program and as the libx264 library used by FFmpeg, HandBrake and most other tools. It is fast, well threaded and keeps fine detail and grain well thanks to its psychovisual optimizations. Because H.264 decodes on practically everything, x264 is still the default choice when compatibility matters more than compression efficiency.

x264 has no numbered releases. Builds are identified by API core and revision, for example core 165 r3222; development now consists mostly of occasional platform and assembly work. Most FFmpeg builds include libx264; standalone binaries are published at artifacts.videolan.org, and the source is at code.videolan.org/videolan/x264.

Encode with x264

Pick the slowest preset you can tolerate, pick a CRF, and adjust from there:

x264 --preset slow --crf 20 -o output.mkv input.y4m
ffmpeg -i input.mkv -c:v libx264 -preset slow -crf 20 -c:a copy output.mkv

The standalone program reads Y4M and raw YUV (plus anything else if built with libavformat or FFMS2 input) and writes raw .264, Matroska, FLV, or MP4 when built with L-SMASH or GPAC. It handles video only, so mux audio and subtitles afterwards. To feed it from FFmpeg, pipe Y4M:

ffmpeg -i input.mkv -f yuv4mpegpipe - | x264 --demuxer y4m --preset slow --crf 20 -o output.264 -

Remux x264’s Matroska output

x264’s built-in Matroska writer does not write a seek index (Cues), so seeking in its .mkv files can be slow or broken. Remux the file (or a raw .264 output) with mkvmerge or FFmpeg (-c copy).

In FFmpeg, -preset, -tune, -crf, -profile:v, -g (keyint) and -b:v map to x264 settings; anything else goes through -x264-params as colon-separated key=value pairs (use , inside values such as psy-rd=1.0,0.15). FFmpeg leaves x264’s own defaults alone unless you set something, so a plain -c:v libx264 encode is preset medium, CRF 23.

Rate control

Mode	Use it for	How
CRF	Almost everything: constant perceived quality, file size follows the content	--crf 20 / -crf 20
Two-pass bitrate	Hitting a file size	--pass 1 then --pass 2 with the same --bitrate
VBV	Streaming, Blu-ray or device limits on peak bitrate	--vbv-maxrate + --vbv-bufsize, with CRF or bitrate
Constant QP	Lossless (--qp 0) and testing only	--qp N

CRF defaults to 23. Lower is higher quality and larger files; ±6 roughly halves or doubles the bitrate. Most encodes land between about 16 (close to transparent) and 23. CRF values are not comparable between encoders, and changing the preset or tune changes the size you get at a given CRF. With 10-bit output the CRF scale extends down to -12.

Two-pass gives the best quality for a target size. x264 runs the first pass with faster analysis settings automatically (--slow-firstpass disables that), and the first-pass output can be thrown away:

x264 --preset slow --pass 1 --bitrate 4000 -o /dev/null input.y4m
x264 --preset slow --pass 2 --bitrate 4000 -o output.mkv input.y4m

With FFmpeg, use -pass 1/-pass 2 with -b:v 4M; the first pass can skip audio and discard its output with -an -f null /dev/null.

VBV limits the peak bitrate a decoder’s buffer has to absorb. --vbv-maxrate is in kbit/s and --vbv-bufsize in kbit, so --vbv-maxrate 10000 --vbv-bufsize 20000 caps the rate at 10 Mbit/s over a 2-second buffer. Combined with CRF (“capped CRF”) it keeps CRF’s quality targeting until a scene would exceed the cap. In FFmpeg the equivalents are -maxrate and -bufsize.

Levels. Without --level, x264 signals whatever level the settings need. At 1080p, --preset veryslow uses 16 reference frames and produces level 5.1, which some TVs and hardware players reject. Set --level 4.1 (or whatever your target device needs) and x264 caps the reference count to fit.

Presets and tunes

Presets run ultrafast, superfast, veryfast, faster, fast, medium (default), slow, slower, veryslow and placebo. Each step trades speed for compression efficiency. veryslow is the usual practical ceiling; placebo is much slower for a negligible gain. Choose a preset first, then tune individual options.

Tunes adjust settings for a type of content and are applied after the preset; your own options override both. Only one of the psy tunes (film, animation, grain, stillimage, psnr, ssim) can be used at a time, though fastdecode and zerolatency can be combined with one.

Tune	Changes (from x264 —fullhelp)	Use for
film	--deblock -1:-1 --psy-rd <unset>:0.15	Live action
animation	--bframes +2 --deblock 1:1 --psy-rd 0.4:<unset> --aq-strength 0.6, doubled refs	Cel animation, flat colours
grain	--aq-strength 0.5 --no-dct-decimate --deadzone-inter 6 --deadzone-intra 6 --deblock -2:-2 --ipratio 1.1 --pbratio 1.1 --psy-rd <unset>:0.25 --qcomp 0.8	Keeping heavy film grain at high bitrates
stillimage	--aq-strength 1.2 --deblock -3:-3 --psy-rd 2.0:0.7	Slideshows, still content
psnr / ssim	Disable psy optimizations (--no-psy; AQ off for psnr, auto-variance AQ for ssim)	Metric benchmarking only
fastdecode	--no-cabac --no-deblock --no-weightb --weightp 0	Weak decoders, very high bitrates
zerolatency	--bframes 0 --force-cfr --no-mbtree --sync-lookahead 0 --sliced-threads --rc-lookahead 0	Real-time streaming, video calls

--tune psnr and --tune ssim make video look worse

They switch off the psychovisual optimizations so the encode scores higher on PSNR or SSIM. The result is blurrier and loses texture, so use them only when you are measuring the encoder against those metrics. For judging quality, look at the video, and see Compare encoding quality for better metrics.

Options worth adjusting

Defaults are for preset medium with no tune. Most people only touch a few of these: a common high-quality starting point for live action is --preset veryslow --tune film --crf 18, adding --aq-mode 3 if dark scenes band. See Psychovisual optimization for the background.

Option	Default	What it does and when to change it
--psy-rd <rd>:<trellis>	1.0:0.0	Psy-RD (first value) favours keeping the source’s energy and detail; psy-trellis (second) does the same in quantization. Film and grain tunes raise psy-trellis to 0.15/0.25. Lower psy-RD (e.g. 0.4 in the animation tune) for flat content. Psy-RD needs --subme 6 or higher; psy-trellis needs trellis.
--aq-mode	1	0 off, 1 variance AQ, 2 auto-variance, 3 auto-variance with a bias toward dark scenes. Mode 3 is a popular choice to reduce banding and blocking in dark areas; it spends more bits on dark scenes.
--aq-strength	1.0	How strongly AQ moves bits between flat and textured areas. Animation tune lowers it to 0.6, grain to 0.5.
--deblock <alpha>:<beta>	0:0	Loop-filter strength offsets. Negative values keep more detail and grain at high bitrates (film -1:-1, grain -2:-2); positive values smooth more (animation 1:1).
--no-mbtree / --qcomp	MB-tree on, qcomp 0.60	MB-tree gives more bits to areas that later frames reference and fewer to areas that quickly disappear. It is a large efficiency gain, but it can flatten grain and dark detail. Raising --qcomp weakens it (MB-tree strength is 5 × (1 − qcomp)); the grain tune uses 0.8. Turning MB-tree off entirely costs a lot of efficiency.
--bframes	3	Maximum consecutive B-frames. 8 (veryslow) is common for high-efficiency encodes; animation benefits from more.
--b-adapt	1	B-frame placement: 0 fixed, 1 fast, 2 optimal. Use 2 (slower and up) with high --bframes.
--ref	3	Reference frames, up to 16. More helps efficiency, especially for animation, but costs speed and raises the required level.
--me	hex	Motion search: dia, hex, umh, esa, tesa. umh (slower and up) is the usual quality choice; esa/tesa are rarely worth the time.
--subme	7	Subpixel motion estimation and mode decision, 0–11. 10 needs --trellis 2 and AQ; 11 disables all early terminations.
--merange	16	Motion search range for umh/esa/tesa. 24 in veryslow; raise it for fast motion or high resolutions.
--rc-lookahead	40	Frames analysed for frame types and MB-tree. 60 in slower and up; more costs memory and speed.
--trellis	1	Rate-distortion optimized quantization: 0 off, 1 final decision, 2 all decisions.
--keyint / --min-keyint	250 / auto	Maximum and minimum keyframe interval in frames. Lower --keyint gives faster seeking and is often set to about 10 × the frame rate.
--open-gop	Off	Lets frames reference across keyframe boundaries for slightly better compression. Leave it off if the stream will be cut or segmented at keyframes.
--threads	auto	1.5 × the logical CPU count by default. x264 scales well, so only lower it to leave CPU for other work.

10-bit, 4:2:2 and 4:4:4

Current x264 builds include 8-bit and 10-bit in one binary (--output-depth 10, or -pix_fmt yuv420p10le in FFmpeg). 10-bit H.264 (High 10 profile) reduces banding and compresses a little better, which made it popular for anime releases, but few hardware decoders, phones, TVs or browsers support it. Use 10-bit x264 only when the video will be played in software players.

FFmpeg 9.0.2 still refuses to write more than 8 bits to Y4M without -strict -1, so a 10-bit pipe looks like this:

ffmpeg -i input.mkv -pix_fmt yuv420p10le -strict -1 -f yuv4mpegpipe - | x264 --demuxer y4m --output-depth 10 --preset slow --crf 20 -o output.mkv -

x264 also encodes 4:0:0, 4:2:2 and 4:4:4 (--output-csp, or the input pixel format in FFmpeg) and RGB (--output-csp rgb, or libx264rgb in FFmpeg). These use the High 4:2:2 and High 4:4:4 Predictive profiles, which have even less decoder support.

Lossless

--qp 0 (-qp 0 in FFmpeg) is lossless and uses the High 4:4:4 Predictive profile. --crf 0 is also lossless with 8-bit output but not with 10-bit, so use --qp 0. Slower presets make smaller lossless files.

Lossless x264 is a very fast intermediate format. --preset ultrafast gives the fastest encoding and decoding, and adding --tune fastdecode speeds decoding further; --preset superfast produces noticeably smaller files and is still very fast. For archival, compare against FFV1, which is designed for that job and has wider support in archival tools.

Complete standalone parameter reference

Every long option in the x264 r3222 command-line parser, grouped as in x264 --fullhelp, with short forms where they exist. Defaults are x264’s own with --preset medium and no tune or profile; presets, tunes and --profile change many of them. Apart from --range, the Video Usability Info options only write metadata and never convert the picture. Optional demuxers, muxers and OpenCL support depend on how your binary was built.

Help

Option	Default	Description
-h, --help	—	Print the basic options and exit.
--longhelp	—	Print the basic and common advanced options and exit.
--fullhelp	—	Print every option, including what each preset and tune changes, and exit.

Presets

Option	Default	Description
--profile	none (lowest profile the settings need)	Restrict output to an H.264 profile: baseline, main, high, high10, high422 or high444. Turns off tools the profile forbids, overriding other options, and fails if the bit depth, chroma format or lossless mode needs a higher profile.
--preset	medium	Speed/compression trade-off: ultrafast, superfast, veryfast, faster, fast, medium, slow, slower, veryslow, placebo. Applied first; any option you set explicitly overrides it. Use the slowest preset you can afford.
--tune	none	Adjust for the source or use case. Psy tunes (pick at most one): film, animation, grain, stillimage, psnr, ssim. Others: fastdecode, zerolatency. Combine with commas, e.g. film,fastdecode.
--slow-firstpass	off	Keep the full analysis settings in --pass 1. By default the first pass switches to --ref 1 --me dia --partitions none --trellis 0 --no-8x8dct and at most --subme 2 for speed. placebo implies it.

Frame-type options

Option	Default	Description
-I, --keyint	250	Maximum distance between IDR frames, in frames; infinite for none. Lower it for faster seeking or to match a streaming segment length.
-i, --min-keyint	auto (keyint/10, at most the frame rate)	Minimum distance between IDR frames. A scene cut closer than this to the previous IDR gets a non-IDR I-frame instead. Clamped to keyint/2+1.
--no-scenecut	off	Disable scene-cut detection, so keyframes are placed only by --keyint. Same as --scenecut 0.
--scenecut	40	Scene-cut sensitivity: higher values insert I/IDR frames at more scene changes. 0 disables detection.
--intra-refresh	off	Replace IDR keyframes with a column of intra blocks that sweeps across the picture over keyint frames. Evens out frame sizes for low-latency streaming; forces --ref 1, and disables --open-gop.
-b, --bframes	3	Maximum consecutive B-frames, 0–16. With adaptive B-frames x264 uses fewer where that is cheaper.
--b-adapt	1	B-frame placement: 0 always use --bframes, 1 fast, 2 optimal (slower, especially with many B-frames).
--b-bias	0	Bias the adaptive B-frame decision toward (positive) or against (negative) B-frames, −90 to 100.
--b-pyramid	normal	Let B-frames be used as references: none, strict (hierarchy allowed by Blu-ray) or normal. Needs at least 2 B-frames.
--open-gop	off	Let frames after a keyframe reference frames before it, with recovery-point SEI marking the random-access point. Slightly more efficient; some players and editors handle open GOPs poorly. Needs B-frames.
--no-cabac	off	Use CAVLC instead of CABAC entropy coding: larger files, faster decoding. Required for Baseline profile.
-r, --ref	3	Maximum reference frames, 1–16. More helps compression with diminishing returns and costs speed and decoder memory. If you set --level but not --ref, x264 lowers it to fit the level’s DPB.
--no-deblock	off	Disable the in-loop deblocking filter.
-f, --deblock	0:0	Deblocking strength and threshold offsets as alpha:beta, each −6 to 6. Negative values keep more fine detail and grain, positive values smooth more. A single number sets both.
--slices	0 (one slice)	Slices per frame. Forces rectangular slices; overridden by --slice-max-size and --slice-max-mbs. Mainly for specs that require slices (Blu-ray, broadcast).
--slices-max	0 (no limit)	Absolute maximum slices per frame; takes priority over --slice-max-size and --slice-max-mbs.
--slice-max-size	0 (off)	Maximum slice size in bytes, including NAL overhead. Used to fit slices into network packets.
--slice-max-mbs	0 (off)	Maximum macroblocks per slice.
--slice-min-mbs	0 (off)	Minimum macroblocks per slice when slices are split by size or macroblock count.
--tff	off	Encode interlaced, top field first. x264 also switches to interlaced coding when the input is flagged interlaced.
--bff	off	Encode interlaced, bottom field first.
--constrained-intra	off	Intra macroblocks may only predict from intra-coded neighbours. Limits error propagation; costs compression.
--pulldown	none	Soft pulldown: code progressive frames and flag repeated fields/frames with pic_struct. Values: none, 22, 32, 64, double, triple, euro; 32 turns 23.976p into 29.97i. Requires CFR input.
--fake-interlaced	off	Flag the stream as interlaced while coding progressive frames. Allows 25p and 30p on Blu-ray, which only permits those rates as interlaced.
--frame-packing	-1 (none)	Stereo 3D frame-packing SEI: 0 checkerboard, 1 column interleave, 2 row interleave, 3 side by side, 4 top-bottom, 5 frame alternation, 6 2D, 7 tile format.

Ratecontrol

Option	Default	Description
-q, --qp	none (CRF mode)	Constant-QP mode. Range 0–69 for 8-bit and 0–81 for 10-bit output (10-bit QPs are 12 higher for the same quantizer step); 0 is lossless. Turns off AQ and MB-tree, so CRF compresses better at the same quality.
-B, --bitrate	none	Average bitrate in kbit/s (ABR). Use it with --pass for a target file size, or with VBV for capped streaming; single-pass ABR without VBV is the weakest mode.
--crf	23.0	Constant rate factor, the default mode: constant quality with variable bitrate. Lower is better quality; +6 roughly halves the bitrate. Range 0–51 for 8-bit and −12–51 for 10-bit output; CRF 0 is lossless only for 8-bit. 17–28 is the usual range.
--rc-lookahead	40	Frames analysed ahead for frame-type decisions, MB-tree and VBV, up to 250. More helps MB-tree slightly but adds memory and latency. Capped at the larger of --keyint and the VBV buffer duration.
--vbv-maxrate	0 (off)	VBV maximum rate in kbit/s; needs --vbv-bufsize. With CRF it caps peaks; set it equal to --bitrate for CBR-like output.
--vbv-bufsize	0 (off)	VBV buffer size in kbit; needs --vbv-maxrate except in ABR mode, where maxrate defaults to the bitrate. Buffer size divided by maxrate is roughly the averaging window in seconds.
--vbv-init	0.9	Initial VBV buffer fill: values up to 1 are a fraction of --vbv-bufsize, larger values are in kbit.
--crf-max	none	With CRF and VBV, the highest rate factor VBV may push to. Stops quality collapsing in hard scenes but can cause VBV underflows.
--qpmin	0	Lowest QP rate control may use. Raising it saves bits on easy content at the cost of quality there.
--qpmax	81 (no cap)	Highest QP rate control may use. The default is clamped to x264’s internal maximum for the bit depth (69 for 8-bit, 81 for 10-bit). Lowering it prevents very coarse frames but can break bitrate or VBV targets.
--qpstep	4	Maximum QP change between consecutive frames, minimum 2.
--ratetol	1.0	How far ABR may drift from the target bitrate while spreading bits (higher = freer distribution, less exact size); also how aggressively VBV avoids underflow. Accepts inf.
--ipratio	1.40	QP factor between I- and P-frames: 1.4 makes I-frames about 3 QP lower.
--pbratio	1.30	QP factor between P- and B-frames. Has no effect with MB-tree on (the default), which sets B-frame QPs itself.
--chroma-qp-offset	0	Chroma QP offset relative to luma, −12 to 12. x264 adds its own adjustment on top: −2 with psy-rd (−1 below 0.25), another −1 or −2 with psy-trellis, and +6 for 4:4:4 with psy on.
--aq-mode	1	Adaptive quantization: 0 off, 1 variance AQ, 2 auto-variance, 3 auto-variance biased toward dark scenes. AQ moves bits from detailed to flat areas, reducing banding and blocking; mode 3 helps dark gradients.
--aq-strength	1.0	AQ strength, 0–3; 0 disables AQ. Tunes set 0.6 (animation), 0.5 (grain) and 1.2 (stillimage).
-p, --pass	none	Multipass: 1 writes the stats file, 2 reads it (final pass), 3 reads and rewrites it (middle passes). Use with --bitrate for an exact file size.
--stats	x264_2pass.log	Stats file for --pass. With MB-tree, x264 also writes <file>.mbtree.
--no-mbtree	off	Disable MB-tree, which lowers QP for blocks that later frames reference. MB-tree is a large compression gain; leave it on.
--qcomp	0.60	Quantizer curve compression, 0–1: 0 is constant bitrate, 1 is constant QP. Also sets MB-tree strength (higher = weaker); 1 disables MB-tree. --tune grain uses 0.8.
--cplxblur	20.0	Temporal smoothing of frame complexity before the QP curve is applied; reduces QP fluctuation in ABR and multipass.
--qblur	0.5	Temporal smoothing of QPs after the curve is applied (multipass).
--zones	none	Per-range overrides, separated by /: start,end,q=N (force QP) or start,end,b=F (bitrate multiplier), optionally followed by more option=value settings such as crf= or deblock=. Frame numbers are inclusive.
--qpfile	none	File of frame type [QP] lines that force frame types (I, i, K, P, B, b) and optional QPs. K is an IDR, or a recovery-point I-frame with --open-gop. Used to put keyframes on chapter or segment boundaries.

Analysis

Option	Default	Description
-A, --partitions	p8x8,b8x8,i8x8,i4x4	Macroblock partitions searched: p8x8, p4x4, b8x8, i8x8, i4x4, none, all. p4x4 needs p8x8; i8x8 needs 8×8 DCT. slower and above use all.
--direct	spatial	Direct motion-vector prediction in B-frames: none, spatial, temporal or auto (picks per frame; used from slow up).
--no-weightb	off	Disable weighted prediction in B-frames.
--weightp	2	Weighted prediction in P-frames, mainly for fades: 0 off, 1 weighted references, 2 also adds duplicate references with different weights. Some players mishandle 2; --bluray-compat limits it to 1.
--me	hex	Integer-pel motion search: dia (fastest), hex, umh (uneven multi-hexagon, used by slower and veryslow), esa (exhaustive), tesa (exhaustive with Hadamard transform). Beyond umh the gains are tiny.
--merange	16	Motion search range in pixels, 4–1024. With dia and hex anything above 16 is ignored; veryslow uses 24 with umh.
--mvrange	-1 (auto)	Maximum vertical motion-vector length in pixels. Auto uses the level’s limit; lower it only for a decoder that needs it.
--mvrange-thread	-1 (auto)	Minimum row buffer between frame threads, in pixels. Limits how far motion vectors may reach into a frame still being encoded; leave on auto.
-m, --subme	7	Subpixel refinement and mode decision, 0–11: 1–5 progressively more quarter-pel search, 6 RD mode decision on I/P-frames, 7 on all frames, 8–9 RD refinement, 10 QP-RD (needs --trellis 2 and AQ), 11 full RD. Psy-RD needs 6 or higher.
--psy-rd	1.0:0.0	Psychovisual strengths as rd:trellis, each 0–10. Psy-RD (needs --subme 6+) keeps detail and grain instead of blurring them; psy-trellis (needs trellis) does the same during quantization. A single number sets psy-RD and zeroes psy-trellis. Tunes: film :0.15, grain :0.25, animation 0.4:, stillimage 2.0:0.7.
--no-psy	off	Disable all psychovisual optimizations that lower PSNR and SSIM, including psy-RD and psy-trellis. Used by --tune psnr and --tune ssim; the result looks worse.
--no-mixed-refs	off	Choose references per macroblock instead of per 8×8 partition. Slightly faster and less efficient; irrelevant with --ref 1.
--no-chroma-me	off	Use only luma in motion search. Slightly faster, slightly worse.
--no-8x8dct	off	Disable the adaptive 8×8 transform, a High-profile tool. Main and Baseline profiles turn it off.
-t, --trellis	1	Trellis (rate-distortion optimized) quantization: 0 off, 1 on the final encode of each macroblock, 2 in all mode decisions (slow and above). Psy-trellis needs it.
--no-fast-pskip	off	Disable early skip detection in P-frames. Slower; can help flat areas and gradients. placebo uses it.
--no-dct-decimate	off	Keep the small coefficients that DCT decimation would zero in inter blocks. Preserves fine detail and grain at a bit cost; used by --tune grain.
--nr	0 (off)	Fast noise reduction in the DCT domain, 0–65536. Values of a few hundred are typical; a real denoising filter does a better job.
--deadzone-inter	21	Inter luma quantization deadzone, 0–32: larger values discard more small coefficients. Only applies where trellis is not used (everything with --trellis 0, mode decisions with --trellis 1). --tune grain sets 6.
--deadzone-intra	11	Intra luma quantization deadzone, 0–32. Same rules as --deadzone-inter.
--cqm	flat	Quantization-matrix preset: flat or jvt (the JVT default matrices). Non-flat matrices need High profile.
--cqmfile	none	Read custom matrices from a JM-format file. Overrides all other --cqm* options.
--cqm4	flat	Set all four 4×4 matrices: 16 comma-separated integers.
--cqm8	flat	Set all 8×8 matrices: 64 comma-separated integers.
--cqm4i, --cqm4p, --cqm8i, --cqm8p	flat	Set intra or inter (p) 4×4 or 8×8 matrices for luma and chroma together: 16 or 64 integers.
--cqm4iy, --cqm4ic, --cqm4py, --cqm4pc	flat	Set one 4×4 matrix (intra/inter × luma/chroma): 16 integers.

Video Usability Info (Annex E)

Option	Default	Description
--overscan	undef	Overscan hint: undef, show or crop.
--videoformat	undef	Analog source format flag: component, pal, ntsc, secam, mac or undef.
--range	auto (from input; pc for RGB)	Signalled sample range: tv (limited) or pc (full). If it differs from the input range, the CLI converts the samples.
--colorprim	undef	Colour primaries flag, e.g. bt709, bt470bg, smpte170m, bt2020.
--transfer	undef	Transfer characteristics flag, e.g. bt709, smpte2084 (PQ), arib-std-b67 (HLG).
--colormatrix	auto (GBR for RGB, otherwise undef)	Matrix coefficients flag, e.g. bt709, bt470bg, smpte170m, bt2020nc.
--chromaloc	0	Chroma sample location type, 0–5. 0 (left) suits most 4:2:0 video; 2 (top-left) is used for BT.2020/UHD.
--mastering-display	none	HDR10 mastering display SEI: G(x,y)B(x,y)R(x,y)WP(x,y)L(max,min), chromaticity in 0.00002 units and luminance in 0.0001 cd/m².
--cll	none	Content light level SEI: MaxCLL,MaxFALL in cd/m².
--alternative-transfer	undef	Alternative transfer characteristics SEI, same values as --transfer. Used for HLG: --transfer bt2020-10 --alternative-transfer arib-std-b67.
--nal-hrd	none	Write HRD parameters and buffering-period/picture-timing SEI: none, vbr or cbr. Requires VBV. Needed for broadcast and Blu-ray; cbr pads with filler and is not allowed in MP4.
--filler	off	Hard CBR: add filler data so the stream never drops below --vbv-maxrate. Implied by --nal-hrd cbr.
--pic-struct	off	Always write pic_struct in picture-timing SEI. Pulldown and interlaced coding write it anyway.
--crop-rect	0,0,0,0	Add left,top,right,bottom to the bitstream cropping rectangle, so decoders hide those pixels. Values must be multiples of the chroma subsampling.

Input/Output

Option	Default	Description
-o, --output	required	Output file. The extension selects the container: .mkv, .flv, .mp4 (needs L-SMASH or GPAC), anything else raw H.264; - writes raw H.264 to stdout.
--muxer	auto	Force the container: auto, raw, mkv, flv or mp4.
--demuxer	auto	Force the input reader: auto, raw, y4m, plus avs, lavf and ffms when compiled in.
--input-fmt	none	Force the libavformat input format (lavf builds only).
--input-csp	i420 (raw)	Pixel format of raw input, e.g. i420, nv12, i422, i444, rgb. Y4M, lavf and ffms inputs carry their own.
--output-csp	i420	Chroma format to encode: i400, i420, i422, i444 or rgb. The CLI converts other inputs to it, so a 4:2:2 or 4:4:4 source is encoded as 4:2:0 unless you set this.
--input-depth	8 (raw)	Bit depth of raw input, 8–16. Y4M carries its own.
--output-depth	8 (10 in a 10-bit-only build)	Encoding bit depth, 8 or 10. Current builds support both in one binary; input of another depth is converted, with dithering when reducing.
--input-range	auto	Range of the input: auto, tv or pc. Overrides what the demuxer reports.
--input-res	required (raw)	Size of raw input as WIDTHxHEIGHT.
--index	none	Index file for the ffms demuxer; written on the first run and reused afterwards.
--sar	from input	Sample (pixel) aspect ratio as W:H, e.g. 1:1 or 40:33. Set it for anamorphic video.
--fps	from input (25 for raw)	Frame rate as a decimal or rational (24000/1001). Forces constant-frame-rate timestamps.
--seek	0	First input frame to encode (0-based).
--frames	all	Maximum number of frames to encode.
--level	auto	H.264 level to signal, e.g. 4.1, 41 or 1b. Without --ref, x264 reduces the reference count to fit; it only warns about VBV and other limits, so set --vbv-maxrate and --vbv-bufsize for real compliance.
--bluray-compat	off	Blu-ray restrictions: at most 3 B-frames and 6 references, strict pyramid, --weightp 1 or lower, AUDs, HRD (vbr), no intra refresh, --min-keyint 1. Combine with the other Blu-ray settings (level, VBV, slices, keyint).
--avcintra-class	0 (off)	Produce AVC-Intra class 50, 100, 200, 300 or 480 (300 and 480 need --avcintra-flavor sony). Requires a matching resolution, frame rate and pixel format, and overrides most settings.
--avcintra-flavor	panasonic	AVC-Intra variant: panasonic or sony (XAVC).
--stitchable	off	Don’t optimize headers for the content, so separately encoded segments can be joined.
-v, --verbose	off	Print statistics for every frame (debug log level).
--no-progress	off	Hide the progress line.
--quiet	off	Suppress all log output.
--log-level	info	Log verbosity: none, error, warning, info or debug.
--psnr	off	Compute and report PSNR. x264 warns that the numbers are invalid unless psy is off and AQ is off (--tune psnr).
--ssim	off	Compute and report SSIM. x264 warns unless psy is off and AQ is on (--tune ssim).
--threads	auto (1.5 × logical CPUs)	Frame-encoding threads, up to 128. Auto is also limited to one thread per two macroblock rows. Very high counts cost a little compression.
--lookahead-threads	auto	Threads for the lookahead; auto derives a count from --threads and the analysis settings.
--sliced-threads	off	Thread by slices within a frame instead of by frames: no extra latency, but less efficient. Used by --tune zerolatency.
--thread-input	auto (on with more than one thread)	Read and decode input in its own thread.
--sync-lookahead	auto (bframes + 1)	Frames buffered between the threaded lookahead and the encoder; 0 disables threaded lookahead for lower latency.
--non-deterministic	off	Let frame threads use slightly more information for marginally better quality, at the cost of run-to-run repeatability.
--cpu-independent	off	Produce identical output on different CPUs by not letting them pick different algorithms.
--asm	auto	Override CPU feature detection with a bitmask or comma-separated names (e.g. sse2,avx2). For debugging.
--no-asm	off	Disable all assembly optimizations. Much slower; for debugging.
--opencl	off	Use OpenCL for the lookahead (OpenCL builds, 8-bit only).
--opencl-clbin	none	File for caching the compiled OpenCL kernel.
--opencl-device	0	OpenCL device index.
--dump-yuv	none	Write reconstructed frames to a raw YUV file, for debugging.
--sps-id	0	ID written into the SPS and PPS, 0–31.
--aud	off	Write access unit delimiters. Some broadcast, transport-stream and Blu-ray targets need them.
--force-cfr	off	Ignore input timestamps and generate constant-frame-rate ones.
--tcfile-in	none	Take timestamps from a Matroska timecode v1 or v2 file for VFR output. Can’t be combined with --fps.
--tcfile-out	none	Write the input timestamps to a timecode v2 file.
--timebase	from input	Timebase as num/den. A single integer is the numerator with --tcfile-in and the denominator otherwise. Not allowed with CFR input.
--dts-compress	off	Remove the initial delay caused by B-frames by shifting DTS in FLV and MP4 output.

Filtering

Option	Default	Description
--vf, --video-filter	none	Input filter chain, filters separated by /: crop:left,top,right,bottom, select_every:step,offset1[,...], and resize (scaling, SAR and pixel-format conversion) in builds with libswscale.

Additional accepted flags and aliases

Option	Default	Description
-V, --version	—	Print the version and build configuration and exit.
--no-b-adapt	off	Same as --b-adapt 0.
--nf	off	Old spelling of --no-deblock.
--filter	0:0	Old spelling of --deblock.
--interlaced	off	Encode interlaced; field order defaults to top field first.
--no-interlaced	off	Encode progressive even if the input is flagged interlaced.
--cabac	on	Use CABAC entropy coding (the default).
--stdout	auto	Old alias of --muxer; takes the same values.
--stdin	auto	Old alias of --demuxer; takes the same values.
--analyse	p8x8,b8x8,i8x8,i4x4	Old spelling of --partitions.
-w, --weightb	on	Weighted prediction in B-frames (the default).
--psy	on	Psychovisual optimizations (the default).
--mixed-refs	on	Per-partition reference selection (the default).
-8, --8x8dct	on	Adaptive 8×8 transform (the default).
--fast-pskip	on	Early P-frame skip detection (the default).
--mbtree	on	MB-tree rate control (the default).
--no-sliced-threads	on (frame threading)	Thread by frames (the default); cancels --sliced-threads.

Complete FFmpeg libx264 private options

The private libx264 options in FFmpeg 9.0.2; the last column names the x264 setting each one sets. A default of -1, auto, unset or INT_MIN means not set: FFmpeg passes nothing and x264’s preset value applies, so -crf -1 does not mean CRF −1. Generic FFmpeg options also reach x264: -b:v (--bitrate), -maxrate and -bufsize (VBV), -g (--keyint), -keyint_min, -bf (--bframes), -refs, -qmin, -qmax, -qcomp, -trellis, -me_range, -subq, -threads and the stream’s colour metadata. -x264-params and -x264opts take x264 library settings, not CLI-only options such as --output or --demuxer.

All libx264 private options

Option	Type	Default	Range / values	Maps to and meaning
-preset	string	medium	string	x264 --preset.
-tune	string	unset	string	x264 --tune; combine with commas, e.g. film,fastdecode.
-profile	string	unset	string	x264 --profile: baseline, main, high, high10, high422, high444.
-fastfirstpass	boolean	true	0 to 1	With -pass 1, use x264’s fast first-pass settings. 0 is the same as x264 --slow-firstpass.
-level	string	unset	string	x264 --level, e.g. 4.1.
-passlogfile	string	unset	string	Same field as -stats. On the ffmpeg command line this name belongs to the generic two-pass prefix option, which sets stats to <prefix>-<stream index>.log (default ffmpeg2pass-0.log).
-wpredp	string	unset	string	x264 --weightp as a string (0, 1, 2). Older duplicate of -weightp, which wins if both are set.
-a53cc	boolean	true	0 to 1	Write A/53 closed captions from the input frames as SEI.
-x264opts	string	unset	string	x264 settings as key=value:key=value; a bare key means key=1. Applied after the other wrapper options. An unknown key or bad value stops the encode.
-crf	float	-1	-1 to FLT_MAX	x264 --crf. Takes priority over -b:v and -qp; ignored with -pass 2.
-crf_max	float	-1	-1 to FLT_MAX	x264 --crf-max. Ignored with -pass 2.
-qp	int	-1	-1 to INT_MAX	x264 --qp (constant QP, 0 = lossless). Used only if -crf is not set; ignored with -pass 2.
-aq-mode	int	-1	-1 to INT_MAX; none=0, variance=1, autovariance=2, autovariance-biased=3	x264 --aq-mode.
-aq-strength	float	-1	-1 to FLT_MAX	x264 --aq-strength.
-psy	boolean	auto	-1 to 1	x264 --psy (0 = --no-psy).
-psy-rd	string	unset	string	x264 --psy-rd, as rd:trellis.
-rc-lookahead	int	-1	-1 to INT_MAX	x264 --rc-lookahead.
-weightb	boolean	auto	-1 to 1	x264 --weightb.
-weightp	int	-1	-1 to INT_MAX; none=0, simple=1, smart=2	x264 --weightp.
-ssim	boolean	auto	-1 to 1	x264 --ssim: report SSIM at the end. For PSNR use the generic -flags +psnr.
-intra-refresh	boolean	auto	-1 to 1	x264 --intra-refresh.
-bluray-compat	boolean	auto	-1 to 1	x264 --bluray-compat; setting it also forces constant-frame-rate timestamps.
-b-bias	int	INT_MIN	INT_MIN to INT_MAX	x264 --b-bias; INT_MIN means not set.
-b-pyramid	int	-1	-1 to INT_MAX; none=0, strict=1, normal=2	x264 --b-pyramid.
-mixed-refs	boolean	auto	-1 to 1	x264 --mixed-refs.
-8x8dct	boolean	auto	-1 to 1	x264 --8x8dct.
-fast-pskip	boolean	auto	-1 to 1	x264 --fast-pskip.
-aud	boolean	auto	-1 to 1	x264 --aud.
-mbtree	boolean	auto	-1 to 1	x264 --mbtree.
-deblock	string	unset	string	x264 --deblock (alpha:beta).
-cplxblur	float	-1	-1 to FLT_MAX	x264 --cplxblur.
-partitions	string	unset	p8x8, p4x4, b8x8, i8x8, i4x4, none, all	x264 --partitions, comma-separated.
-direct-pred	int	-1	-1 to INT_MAX; none=0, spatial=1, temporal=2, auto=3	x264 --direct.
-slice-max-size	int	-1	-1 to INT_MAX	x264 --slice-max-size, in bytes.
-stats	string	unset	string	x264 --stats. The ffmpeg CLI fills it in from -passlogfile when you use -pass.
-nal-hrd	int	-1	-1 to INT_MAX; none=0, vbr=1, cbr=2	x264 --nal-hrd; needs -maxrate and -bufsize. cbr is not allowed in MP4.
-avcintra-class	int	-1	-1 to 480	x264 --avcintra-class; 300 and 480 also select the Sony flavor.
-me_method	int	-1	-1 to 4; dia=0, hex=1, umh=2, esa=3, tesa=4	x264 --me; same field as -motion-est.
-motion-est	int	-1	-1 to 4; dia=0, hex=1, umh=2, esa=3, tesa=4	x264 --me.
-forced-idr	boolean	false	-1 to 1	Code forced keyframes (e.g. from -force_key_frames) as IDR frames. Otherwise they are x264 keyframes, which are non-IDR recovery points with open GOP.
-coder	int	default	-1 to 1; default=-1, cavlc=0, cabac=1, vlc=0, ac=1	Entropy coder: cabac/ac or cavlc/vlc. default keeps x264’s choice (CABAC unless the profile or tune disables it).
-b_strategy	int	-1	-1 to 2	x264 --b-adapt (0, 1, 2).
-chromaoffset	int	0	INT_MIN to INT_MAX	x264 --chroma-qp-offset; 0 leaves x264’s value.
-sc_threshold	int	-1	INT_MIN to INT_MAX	x264 --scenecut; 0 disables scene-cut detection.
-noise_reduction	int	-1	INT_MIN to INT_MAX	x264 --nr.
-udu_sei	boolean	false	0 to 1	Write user-data-unregistered SEI from the input frames’ side data.
-x264-params	dictionary	unset	key=value list	x264 settings as key=value:key=value, using x264’s long option names without -- (e.g. no-mbtree=1). Applied last, so it overrides everything else. An invalid entry only logs a warning and is skipped.
-mb_info	boolean	false	0 to 1	Pass per-macroblock hints (AVVideoHint side data) to x264. Only useful for applications using the API.

Related pages

• H.264 / AVC
• x265
• FFmpeg
• MKVToolNix
• Psychovisual optimization