Frequently Asked Questions

A quality scale tells the encoder how much error it may accept before it spends more bits. Low values on a constant-quality scale mean a tighter tolerance and a larger file; high values mean a looser tolerance and a smaller file. The scale is not a percentage and is not comparable across codecs, so a value tuned for one encoder says nothing about another. What matters is that the encoder targets a perceived quality rather than a fixed size.

Use an average bitrate when the delivery channel imposes a hard budget: a fixed streaming profile, a disc-free archive with a size ceiling, or a link whose throughput is known. Use a quality target when the goal is consistent visual result and the file size may float. Mixing the two is possible with a capped quality mode, which keeps the quality target but refuses to exceed a stated ceiling.

A slower preset gives the encoder more time to search for cheaper ways to represent the same picture. It examines more candidate motion vectors, tries more block splits and tests more reference frames. The result is usually the same visual quality at a lower bitrate, not a different quality at the same bitrate. The cost is linear or worse in time, so the useful range is bounded by patience and by the number of jobs waiting.

Two-pass encoding helps when a bitrate target must be met precisely. The first pass measures how complex each scene is; the second distributes the budget accordingly, so quiet scenes do not waste bits that action scenes need. If a constant-quality target is used instead, the second pass adds little, because the encoder already adapts scene by scene. Modern quality modes make two-pass mainly a size-control tool.

Drift usually begins with a timestamp problem, not an encoder problem. Variable frame rate sources, trimmed clips and streams with a non-zero start time can all carry timestamps that the output stage reinterprets. When the video and audio timelines are reconstructed from different assumptions, the tracks separate progressively. Forcing a constant frame rate, or passing timestamps through untouched, removes most of these cases.

A mixdown rebuilds the audio into a new channel layout, for example folding six channels into two. A pass-through copies the encoded audio stream into the new container without decoding it. Pass-through is lossless and cheap but only works when the target container accepts the original codec. A mixdown changes the data, so it costs time and, for lossy codecs, a second generation of quality loss.

Normalise them when the output must satisfy a fixed delivery specification. A player will resample mismatched rates on the fly, but the result depends on the player, and some hardware decoders refuse unusual layouts outright. Setting one sample rate and one channel layout across a batch removes an entire class of playback surprises, at the cost of one resampling step.

Choose MP4 when the file must play on consumer hardware, mobile devices and browser players, since support is universal. Choose Matroska when the goal is archival or when the content needs stream types that MP4 handles awkwardly, such as many subtitle tracks or uncommon audio codecs. The container does not change the encoded picture or sound; it only changes how tracks are labelled, indexed and seeked.

Hardware acceleration moves encoding onto a fixed-function block, which is far faster and much lighter on the processor. The trade-off is a smaller search space and coarser rate control, so a hardware encoder at the same quality target usually produces a larger file than a software encoder at a slow preset. It is the right choice for real-time capture and bulk conversion, and the wrong choice for a final archival master.

Verify the output structure rather than the pictures. Inspect the container to confirm the expected streams, codecs, resolutions and durations; check that the reported duration matches the source within a small tolerance; and scan the log for dropped frames or encoder warnings. A short visual sample from the middle of the file catches the rare case where the structure is correct but the content is wrong.