Hacker News Reader

Comment by Wowfunhappy | Hacker News Reader

One thing jumps out to me in the article: they're recommending crf 30 as a starting point for the x264 encode, and they suggest experimenting with values as high as 40.

Unless there's some detail I'm missing... crf 30 is going to look awful for anything ≤ 1080p, and I can't think of a time I'd ever use a value that high. Maybe for 4K content it would be alright, but I don't think that was specified. 40 is just ridiculous.

(I have no idea what the crf scale is like for x265 or av1, so no comment there.)

View on HN

You're correct. An x264 CRF of 30 would look pretty miserable on almost anything compared to the source.

The article also uses the Main profile in that `-crf 30` example, negating quality even more. The author writes that this must be done to enable compatibility with Safari, but AFAIK Safari has supported H.264@High 5.1 for a while now. Using High profile would help reduce some of the compression artifacts that will result from using a CRF of 30.

EDIT: Looking at this "Evil Martians" webpage, it appears this blog post is just marketing for some generalized programming contracting company. I doubt they have actual video experts under their employ - these guys are just web devs who know enough bash to use FFmpeg from the CLI. I doubt they understand the implications of the commands they're running.

> I doubt they understand the implications of the commands they're running.

I put a description for every arguments used by me in FFmpeg commands.

https://evilmartians.com/chronicles/better-web-video-with-av...

Is it not enough?

Not really, no.

Encode quality has become a science:

https://medium.com/netflix-techblog/dynamic-optimizer-a-perc...

For a deeper dive on FFmpeg options you can crib from the ‘scene’ community:

https://scenerules.org/t.html?id=tvx2642k16.nfo

Or grab Handbrake, pick some built-in device or streaming target, and see the tuned default CLI options.

This is great links, thanks.

What arguments can be added to FFmpeg AV1 encoding command to improve the result for browsers?

Was this dynamic optimizer framework from Netflix released already to be used in web video encoding for webpages?

It has basically been the case for every AV1 advocate to compare about video codec using commands that doesn't make sense and benchmarks that were meaningless. Actually that is not strictly true because this happen with every On2 codec prior to their acquisition from Google.

You could also get much higher VMAF scores if you put some time into x264. x264 was never tuned for VMAF, I wish it had a mode just like SSIM and PSNR. But then those were done during early days just to prove a point. I doubt anyone on x264 cares about it.

As much as I like VMAF, and I thought we finally made something with Big Data and Machine Learning so we don't have to rely on PSNR and SSIM, VMAF it is still far from perfect. But my guess Netflix will constantly refine the model and add more improvement to it.

You are correct. CRF 23-27 is a sane range for good quality and compression. Once you hit 30 there are too many artifacts and it looks bad. Not sure why the author chose to recommend that for x264.

Good idea. I will change the number in the article.

Reply on news.ycombinator.com