BEFORE — One fixed bitrate ladder for every title

Netflix doesn't encode every show at the same rate. They run a quality model on every single title and pick a custom recipe just for that video. And for most of streaming, this was a solved problem. You pick a bit rate ladder, you run it, every show gets the same settings. And then around 2015, Netflix said, nope, we're analyzing every piece of content individually. And the piece of video engineering that nobody outside of Netflix thinks about is an action movie with fast motion, complex backgrounds, and a Pixar cartoon with flat colors and smooth gradients compresses very well and you can tell that nobody can tell the difference at a low bit rate. But live action with complex backgrounds needs way more bits to look the same. A fixed ladder is either overspending on the easy stuff or underspending on the hard stuff or both. So Netflix built a pipeline where they encode every new video at many bit rates and resolutions, and then they run a quality metric on each one called VMAF, which they developed and open-sourced. They then find the sweet spot on the quality vs. bit rate curve and build a custom ladder for that title. Sometimes 20% less bandwidth, often better picture quality for every show on the platform. And the pattern generalizes. Anywhere you have a setting that's currently global, but the right value clearly depends on the specific input, there's a per-something optimization sitting there. The clever part isn't the encoder itself. It's the decision to measure first and then pick second. THE PAYOFF — Custom ladder wins on both axes. ~20% less bandwidth, often better picture.