Encoder Presets: Trading Time for File Size
Compressing the same video with the fastest preset takes five minutes; the balanced one takes twenty. That extra time is not wasted waiting — the encoder spends it searching for smarter ways to write the same picture.
What a preset actually does
Video encoding is fundamentally about describing a picture with as few bytes as possible. The encoder can take shortcuts (fast, but wasteful) or run exhaustive motion search and refined quantization decisions (slow, but compact). The preset is that effort dial: ultrafast does the bare minimum, veryfast adds the common optimizations, medium layers on deeper analysis.
The key takeaway: presets barely affect whether a file plays — they affect how much quality fits into a given size. Medium typically produces files 15-30% smaller than ultrafast at equal perceived quality. The extra minutes are converted directly into compression efficiency.
Choosing between the three tiers
Veryfast (the recommended tier) is the sweet spot of efficiency versus time: for everyday clips it encodes at roughly 1-3x video duration in a browser, with quality differences from slower tiers that are essentially invisible. Use it for daily compression, social uploads, and quick conversions.
Ultrafast fits when you simply cannot wait: long source footage, older machines, or a quick sanity check of your parameters — draft with ultrafast, then render the final version at veryfast. Medium is the archival tier for one-off encodes where time is irrelevant and every megabyte counts; medium plus CRF 24 is a classic archive combo. Slower tiers past medium bring diminishing returns, so web tools generally stop there.
Preset and CRF are different dials
CRF sets the quality target; the preset decides how much effort goes into hitting it. CRF 26 at medium and CRF 26 at ultrafast aim for the same quality — but medium reaches it with fewer bytes.
The correct tuning order is: pick CRF first (your quality requirement), then preset (your time budget). Going the other way leads to the classic loop of endlessly nudging CRF to compensate for a mismatched preset. The same logic applies when targeting a file size: estimate the bitrate first, lock CRF, and let your available time decide the preset.