BitrateCraft · Video pipeline · bitratecraft.org
Video Quality: RF, Presets and Tune
Quality in HandBrake is expressed as a rate factor, conventionally written RF. It is a perceptual target rather than a bitrate ceiling: the encoder keeps spending bits until the frame reaches the requested quality, then stops. A calm interview and a fast camera pan are therefore judged on their own terms, and one number can serve a whole library without constant re-tuning. This page sets out a working model of the RF scale, the preset ladder that trades encoding time for compression efficiency, the tune options that bias encoder decisions, and the profile constraints applied by software and hardware encoders alike.
Reading the RF scale
Lower values mean higher quality and a larger file; higher values mean the opposite. The mapping between the number and the resulting bitrate is steep and non-linear, close to a factor of two for every six points, so a small move at the low end costs far more than the same move higher up the scale. Treat it as a dial with a handful of well-marked detents rather than a continuous instrument.
| RF | Intent | Observed behaviour |
|---|---|---|
| 14–16 | Archive of grainy or difficult source | Very large output, fine detail and noise preserved, slowest runs |
| 18–19 | High-quality master for a personal collection | Texture and foliage retained, size roughly doubled against 21 |
| 20–21 | General purpose | Balanced compromise, artefacts rare on typical displays |
| 22–23 | Space-constrained library | Minor softening in motion, banding possible in smooth gradients |
| 24–28 | Preview or screening copy | Visible blocking on fast motion, thin fine detail |
Two habits keep the scale honest. Encode a few short, representative segments before committing to a full run, and judge the result at the display size you actually use. Artefacts that are obvious on a large panel can disappear entirely on a phone.
The preset ladder
A preset controls how hard the encoder works, not the target quality. Moving down the ladder spends more compute to reach a given RF with fewer bits, and the bill is paid in wall-clock time. HandBrake exposes the familiar ordering from ultrafast through placebo.
- veryfast to fast — suitable when storage is not tight and turnaround matters.
- medium — the usual starting point, a reasonable balance for most libraries.
- slow to slower — measurable efficiency gains on detail-rich sources.
- veryslow and placebo — diminishing returns, useful mainly as a reference point.
The practical rule is to choose the slowest preset whose run time still fits your schedule, then let RF control size. Doubling the depth of the preset typically buys a few percent of bitrate at equal quality; moving the RF a step changes size far more dramatically.
Tune options
A tune nudges the encoder's internal decisions for a class of material. It does not move the RF target, but it changes what the available bits are spent on.
- film — a general live-action bias and the safe choice for mixed content.
- animation — favours flat regions and clean edges, useful for drawn material.
- grain — preserves film grain at the cost of bitrate; pair it with a lower RF.
- stillimage — optimised for near-static frames.
- fastdecode — limits features that slow playback on weak hardware.
- zerolatency — removes look-ahead buffering, intended for capture rather than storage.
Speed, quality and profile
Software encoders expose profiles such as baseline, main and high, together with levels that cap resolution, bitrate and reference-frame counts. Profiles matter for device compatibility rather than for quality directly; at a given RF a high profile generally uses fewer bits than a constrained one. Hardware encoders present a different surface. The same RF field maps onto the vendor's internal constant-quality mode, while the vendor's own numbered preset governs speed. Hardware paths are quick and power-efficient, but they usually need a slightly lower RF to match a software encode of equal visual quality, and they degrade sooner on very grainy material.
When comparing software against hardware output, hold the RF constant, change one variable at a time, and judge on identical displays. Reusing one RF across both families without adjustment is the most common source of disappointing comparisons.
Common questions
Is RF the same as a bitrate target? No. RF asks for a quality level and lets the bitrate float; average-bitrate mode fixes the size and lets quality float.
Should the preset match the RF? They are independent. The preset decides how efficiently a given RF is reached, not how high the target sits.
Why does the second pass of an average-bitrate encode look different? Because it is a size constraint, not a quality constraint; move to RF if consistency across scenes matters more than a predictable file size.
