Choose quality controls, bitrate, presets, and passes
Rate control answers a product constraint. Choose it only after deciding which outcome must remain stable.
| Requirement | Appropriate starting mode | Main tradeoff |
|---|---|---|
| Similar perceptual quality across varied scenes | CRF/constant-quality | Output size is unknown |
| Delivery or storage target over a whole program | VBR, often multi-pass | Quality varies to meet the target |
| Tight instantaneous transmission bound | Constrained quality, VBV, or CBR | May sacrifice quality in complex scenes |
| Diagnostic experiment | Constant quantizer | Usually not a delivery recommendation |
CRF and presets
Section titled “CRF and presets”CRF values are encoder-specific. A value of 30 in SVT-AV1 does not mean the same thing as 30 in x264. Likewise, preset scales differ. Within one encoder, test several values on the same source segment and record output size, encode time, objective metrics, and visual findings.
One pass or multiple passes
Section titled “One pass or multiple passes”For a quality-targeted encode, start with one pass unless the encoder’s current documentation gives a specific reason to use more. For a required average bitrate or file size, multi-pass VBR can allocate bits using information gathered across the full program. SVT-AV1 documents multi-pass as most useful for VBR targets and less critical for ordinary CRF work.
Reproducible comparison
Section titled “Reproducible comparison”Keep all of these fixed while testing one control:
- exact source segment and stream mapping;
- scaling, cropping, filters, frame rate, pixel format, and color pipeline;
- encoder build/version and thread constraints;
- audio and subtitle handling; and
- metric implementation plus viewing conditions.
Never keep a setting merely because it won a single aggregate score. Inspect outliers and visually important scenes.