Processing Order: Crop, Scale, Compress and Captions
Crop, scale, watermark, burn captions, compress — most of these can run in any order and still finish, but the quality differs a lot. Order is not superstition; it is the dependency of what each step builds on.
Iron rule: compression goes last
Compression is lossy encoding: every pass costs one irreversible quality tax, and any operation performed on a compressed file re-encodes it again — two taxes total. In the correct pipeline compression is the terminus: every earlier step runs on lossless or lightly-touched material.
Merge intermediate steps when possible: crop-and-scale in one pass beats two passes (each re-encode is a generation loss). When chaining operations, ask one question: "can this step take the previous step output as input, instead of a compressed artifact?"
Framing actions go first
Crop, rotate, and mirror rearrange pixels — they are the foundation every later step stands on. Fix the composition first, and only then do caption placement and scale targets make sense. Cropping first has a practical bonus: cropped-away pixels never enter later encodes, so compression runs faster and smaller.
Scaling follows cropping: define the content area, then scale to the target resolution — the scaled pixels are all final pixels. The reverse order wastes encoding effort on pixels that get cropped away anyway. Resolution unification for mixed sources belongs in this stage too.
Where captions and watermarks sit
Overlaid elements — captions, watermarks, text — go after composition, before compression: they need full clarity, and stacking them after compression means they get compressed themselves. After composition means they can be laid out precisely to the final frame (a later crop would cut misplaced captions).
A reference order: crop/rotate, scale/unify, captions and watermarks, compress. Exception: sidecar subtitles skip the pipeline entirely — they ship alongside the video and stay switchable; only burned captions occupy the pre-compression slot.