Anatomy of a preset
A preset is one word that stands for a list of settings. x264 and x265 use the same ten words, from ultrafast to placebo, and each word means a different list in each encoder. This page opens the presets up: how far each one moves from medium, and which assignments separate any two.
Compare two presets
Choose an encoder and two presets. The table lists every setting that has a different value in the two, with both values. For x264 the settings are named as on the page about presets as search pruning; for x265 they are its option names.
These are the values each preset assigns on top of the defaults. Tunings, your own options, profiles, pass settings and validation can change the configuration that runs. x264's ultrafast keeps "spatial" for direct prediction here, the default it inherits; it also sets B frames to zero, and validation then switches direct prediction off. "None" for extra partition analyses means the optional analyses are off: 16×16 prediction, intra included, is still available.
| Setting |
|---|
param_apply_preset and x265's x265_param_default_preset, on top of each encoder's defaults. x265's numbers are its own scales: its --subme, --rd and --b-adapt values do not mean what x264's do. The x265 table has the 33 fields whose values differ across its ten presets; an assignment that only repeats a default, as the fastest presets make for hevc-aq and the quantization group size, adds no row. x264's 20 rows group the 21 fields its preset code assigns: the two partition fields share a row. The decision level, --subme, sets motion refinement as well as decision methods, and the trellis values name x264's option levels 1 and 2.How far each preset is from medium
Each bar is a share of that encoder's own list: 17 of 20 is a longer bar than 19 of 33.
- x264's veryslow differs from medium in 10 of the 20 settings, and its ultrafast in 17. The fast end is further from the defaults than the slow end.
- x265's veryslow differs from medium in 17 of 33, its slower in 18 and its ultrafast in 19. A count treats every field alike, so a larger one is not more work.
- One step can be small. x264's slow changes 5 settings from medium; x265's slow changes 13.
- End to end, nearly everything moves. From ultrafast to placebo all 20 of x264's settings differ, and 30 of x265's 33.
What a step is made of
- Search effort. More reference frames, a wider and more thorough motion search, more partition shapes, deeper transform trees. These dominate the slow side in both encoders.
- Shortcuts. x264's fast skip test stays on until placebo. x265's early skip is off in fast and from slow onward, and its limits on references, modes and transform search change value at many steps.
- Lookahead. It generally grows toward the slower presets, with repeated values: 60 in x264 from slower on, 15 in x265 from veryfast to fast. x264's two fastest presets drop it to zero.
- Whole tools, only at the fast end. x264's ultrafast turns off the arithmetic coder, the deblocking filter, the 8×8 transform and adaptive quantization. Those change the stream a decoder sees, not just how hard the encoder looks.
Trade-offs
- A preset does not choose the rate factor or bitrate target. Those are rate control options you set. Presets do change coding tools, the lookahead and how quality is allocated, through adaptive quantization in both encoders and macroblock-tree in x264, so a fixed rate factor can give different sizes and quality at different presets. Compare presets with curves.
- The same word is not the same effort. x264's slow and x265's slow are different lists on different scales, and nothing here measures either.
- Options given with a preset override it. A preset is a starting point, applied before tunings and before your own options.
- This page does not compare AV1 speed settings. libaom's speed setting and SVT-AV1's preset are numbers used throughout their code, and no comparable table was extracted for this course.
Sources
Read 2026-11-15, at the commits linked. The validation of direct prediction, the partition flags and the two repeated x265 assignments were read 2026-11-16.
- x264 at commit c24e06c (GitHub mirror):
param_apply_presetand the defaults incommon/base.c, as read for the page on presets as search pruning; direct prediction switched off with no B frames inencoder/encoder.c; the partition analysis flags inx264.h. - x265 at commit 4191822:
x265_param_default_presetand the defaults insource/common/param.cpp. The table was extracted from the assignments in each preset's branch. - Not checked: what x265's settings do beyond their names, except those explained on other pages; the encoding time or compression of any preset; what x265's
hevc-aqswitch does: its default and the assignment in ultrafast are the same value, as are those of the quantization group size; what validation does to any x265 preset; tunings; libaom's and SVT-AV1's speed settings.