Per-title encoding gives every asset its own bitrate ladder instead of running the whole catalog through one fixed list of resolution-bitrate rungs. The mental flip is to stop fixing the bitrate and letting quality float, and instead fix a quality target (usually a VMAF score) and let each title's bitrate land wherever it must to hit that score. In a pipeline you encode a measurement grid per asset, score each encode against the source, and read the ladder off the result: an easy cartoon may reach VMAF 95 at about 2 Mbps while a grainy sports clip needs roughly 7.5 Mbps for the same target. The quality target is the input and the bitrate is the output. The key catch is that a single VMAF score hides bad shots in its mean, so check the low percentile. It is the same idea as the convex hull (which picks each rung's resolution) and per-shot encoding pushes it finer.
