BitrateCraft · Presets and operations · bitratecraft.org

Batch Queue and Automation Workflows

A single encode is a task; a library is a workflow. Once more than a handful of titles must pass through the same pipeline, the problem shifts from encoder settings to job orchestration: ordering, exceptions, and the ability to leave the machine alone and trust the result. This page covers queue ordering, add-all behaviour, per-title overrides, watch folders, and the discipline of repeatable job sets.

Queue ordering and add-all

HandBrake builds a queue from whatever is loaded, and the order in which titles are enqueued usually follows the order they were opened. That is fine for two files and wrong for four hundred. Two effects matter. First, the encoder processes the queue serially, so the position of a long title determines how much work precedes the short ones. Second, a hardware encoder that holds a fixed-function block benefits from keeping the queue resident while a software encode would rather be scheduled around interactive use.

The add all action scans the source and creates one job per eligible title. On a disc structure this can produce dozens of jobs, most of them short menus or fragments. On a folder of files it produces one job per file. Before queueing a large add-all, open the title list, sort by duration, and remove anything under a minute. A ten-second clip encoded at high quality wastes the same overhead as a feature.

Per-title overrides

Blanket settings are a starting point, not a policy. Some sources need a different crop, a different deinterlace decision, or a higher quality value. The queue stores settings per job, so an override applied to one title does not disturb the others. Build a small matrix of cases and apply it deliberately:

Animation and flat-colour material: raise the quality value, enable a gentle denoise.
Grain-heavy film: keep the quality value high and avoid denoise, which smears texture.
Interlaced broadcast captures: set the deinterlace filter before queueing; detection is not automatic per title.
Silent or mono sources: drop to a single audio track to save space without audible loss.

Storage drive used as the destination volume for a batch run — Destination capacity and free-space headroom should be checked before a long unattended queue, not during it.

Watch folders

HandBrake itself does not watch a directory. Automation is assembled around it: a script polls a drop folder, and for each new file found it launches an encode with a fixed preset, then moves the source aside. The pattern is simple and robust. Three rules keep it reliable: write incoming files to a temporary name and rename only when the copy is complete, so a half-written file is never picked up; log every action with a timestamp; and move rather than delete the source so a failed job can be replayed.

Set the poll interval to a value that suits the source size. Too aggressive and a large copy is seen mid-flight; too lazy and the folder backs up. A one-minute interval is a reasonable default for media arriving over a network share.

Repeatable job sets

A job set is a named collection of presets and overrides that can be reproduced from the command line or from a saved queue export. The value is reproducibility: the same inputs and the same set produce the same outputs, which makes verification meaningful. Export the queue before a large run, keep the export beside the sources, and treat it as part of the deliverable.

Version the recipe, not just the result. A preset saved next to the encoded file explains how the file was made months later, when memory has faded.

Unattended processing

Confirm free space on the destination is at least twice the estimated output.
Disable sleep and hibernate for the duration; a suspended machine stalls the queue.
Set a power profile that favours sustained clocks over burst, to avoid thermal surprises.
Send queue progress to a log file, and have the log tailed so a stalled job is visible.
Collect results and verify the first and last job by hand before trusting the whole run.

The goal is a queue that can be trusted to finish overnight. That trust comes from explicit ordering, deliberate overrides, and logging that survives until the morning.