How to 2-pass h264 with CRF and maximum bitrate with ffmpeg?

Question

Based on CRF Guide (Constant Rate Factor in x264, x265 and libvpx), quoting:

Streaming nowadays is done a little more cleverly. YouTube or Netflix are using 2-pass or even 3-pass algorithms, where in the latter, a CRF encode for a given source determines the best bitrate at which to 2-pass encode your stream. They can make sure that enough bitrate is reserved for complex scenes while not exceeding your bandwidth.

also the article Saving on Encoding and Streaming: Deploy Capped CRF also claims that "Capped CRF" is the highest quality while also making sure bitrate does not go above X but they propose a single pass with:

ffmpeg -i input_file -crf 23 -maxrate 6750k -bufsize 6750k output_file

What's the correct way to generate the highest quality possible file inside a bitrate constraint using ffmpeg with x264?

Answer 1

What's the correct way to generate the highest quality possible file inside a bitrate constraint using ffmpeg with x264?

There is no "correct" way. This is still an evolving area of research, and about half the articles you have read on the subject are more marketing that they are science. The "correct" way may also be different for every video. Netflix for example just encodes the video a couple dozen ways, and keeps the one it thinks is best and throws the rest away.

The closest thing to a "best" single answer, would be 2 pass placebo capped CRF.

Answer 2

I'm not sure you gathered the spirit of that article as intended. The author's point was to describe "capping" or constraining the CRF setting by creating a bandwidth ceiling per frame. The purpose of which is to constrain/limit the overall file size to a pre-defined maximum. This forcibly limits the freedom of the CRF algorithm with respect to its upper limit of bits-per-frame.

Now, take that concept and think about applying ffmpeg's 2-pass model to it. Running 2-pass on capped CRF (aka Constrained CRF) is redundant and a waste of time. First, how would you constitute the 2nd pass? If you repeat the first pass logic, you'll likely attain a smaller target file, but you'll also likely dilute the overall image quality to some extent. The target file size will still have the same maximum possible size it had before. The 2nd pass will almost undoubtedly take longer to process than the first pass, because of how the 2-pass process works in ffmpeg, and the fact you'll be compressing already compressed images again.

Overall, it's just plain a bad idea. It's a time waster and you'd likely end up with worse quality versus the first pass result. Best case scenario, they will look the same. If file size is your top priority, you need to move away from H.264 and choose a more recent codec library optimized for compression (such as h.265/HEVC or VP9, possibly AV1). Constrained CRF is not a poor idea at all, but attempting to run 2-pass encoding with it is.