Ffmpeg 5.0 | Hacker News Reader

Ffmpeg 5.0

ffmpeg.org

115 pointsLeoPanthera32 comments

https://news.ycombinator.com/item?id=29939789

FFmpeg 5.0 (https://github.com/FFmpeg/FFmpeg/blob/release/5.0/Changelog)

113 points|_Gyan_|3 days ago|27 comments

Previously: https://news.ycombinator.com/item?id=29939789

I keep being impressed by this project. Has served me so well over the years, and the depth of the technologies is so exciting. Assembling ffmpeg command line arguments is a meditative exercise, with almost arbitrarily deep research allowed.

I do wish there was an official complementary UI that would make all the options explicit.

I would be very afraid of what that UI would look like. I have a feeling it would be more difficult than a Sony camera menu system.

They exist for Windows (bulk ffmpeg encoder is popular) and they look horrible, probably because they just have to.

Handbrake is the best frontend for it and "works fine" on basic settings and looks OK. But try to do anything advanced and....

Video processing is just really, really complicated when it involves compression/transcoding. There's no other way around it. And that's alright!

Yeah, I think a more workable "interface" to the complexities of FFmpeg would be some sort of terminal "shell" application that allows the user to easily tab-complete their command and visually separate out the command's various parts (i.e., input opts and output opts on separate parts of screen; each track in the container tabbed out and line-returned from the last track; etc.)

Also some users may benefit from a program that can identify what the particulars are about the input file(s) to help them make better commands. Or even ask them first what kind of result they want and work backwards from there. If the user knows they want multiple PNGs and then the input file is an MPEG-4 container with an x265 video bytestream, it's pretty clear what the user may want to do. But unless you're quite initiated, today's FFmpeg still really makes the average user work for it.

> Yeah, I think a more workable "interface" to the complexities of FFmpeg would be some sort of terminal "shell" application [...]

I was thinking something similar, but hadn't thought about tab completion - interesting idea! I was picturing something with a nano-like interface. Run the program from your terminal, get a simple set of menus you can navigate with a keyboard, tick off boxes for options, etc., then have an auto-updating command example output in a bottom "pane" showing you in real time what the cli args would be for what you specified.

In C* land I think they use ncurses to build those, if memory serves, but I probably wouldn't write it in C myself just since I'm not very well versed in it. Probably Go or Rust...

Anyway, good idea. If I hack on this I may see about implementing something like that, some day!

VLC is pretty close

Perhaps as an interface to `ffplay`, but it's certainly not complete enough to be the GUI for `ffmpeg`.

I always thought ffmpeg was a massive, impossible-to-use library that solved all cases under the sun.

Well regardless of the latter, I learned the former was not true as I was looking for a way to turn an mp4 into a gif without some silly desktop app.

    ffmpeg -i in.mp4 out.gif

That's it! Incredibly easy to use after all.

FFMPEG comes with a lot of training wheels, and it tends to make the right "guesses" the majority of the time. That is meant as a HUGE compliment.

Edit: maybe not training wheels. more like gutter bumpers in bowling. no matter what you do, you're going to hit some pins (make a playable output). with just a little trying, you can easily roll strikes

It's incredibly easy to use as long as you are happy with the default settings. The hard part comes in when you want to apply some sort of transformation but you don't know what a lot of the technical terms mean.

Maybe it should come with a sticker "No user servicable parts inside". If you don't know what you're doing, don't take the cover off and start poking with metal objects. /s

There's no better way to learn than to face something you don't know. 99/100 things you don't know or can't figure out, someone on the internet as been there done that. Come across enough of these things, and you start to think like ffmpeg!

Ffmpeg is one of those things that I repeatedly find replaces other tools. It can convert video, it can deinterlace it, it can apply effect filters, etc.

It’s one of those products that has effectively replaced everything else I use video/audio related because ffmpeg is free and does everything.

I've had trouble getting it to properly handle transparency when outputting animated GIFs. The transparent areas wouldn't get cleared between frames without some extra settings. Also you need a special set of transforms to make it generate and use a color palette.

And the APNG support was even worse. I spent days trying to get it to output an APNG that was within Discord's sticker size limits without losing quality. Eventually gave up on it and switched to pngasm which easily produced a small enough output.

EDIT: I should clarify that I still think ffmpeg is amazing, but I find that it can be tricky and does have some limitations that seem to come from being the input and output formats being so flexible.

Simple use cases are simple to solve. It gets hairy quite quickly for exotic formats or if you really want to tune the encoder. It does solve all cases under the sun, though, it is quite impressive.

>I learned the former was not true

It is true, but it applies to the library (libav...). As someone who has tried to use it I can confirm that it is possible to use correctly without a lot of work and access to weirdly encoded files to test your code. They do not make it easy to write correct come.

possibile -> not possible

Getting really good quality gifs out of it, on the other hand, requires two passes and some arcane command line options (or used to anyway) - though it is at least possible and produces impressive results given the limitations of the format.

It is like the imagemagick of video

I'm perpetually impressed by the value FFmpeg adds to society. I imagine an accurate tally of all the places it's used would be impossible, but I can't help but imagine it would be eye-opening in it's breadth.

Two noteworthy items that stand out to me:

> AV1 Low overhead bitstream format muxer

> speex decoder

Congrats on the 5.0 milestone to all contributors!

On a current project I needed to get RMS volume of the mids of an audio file in 1-second isolated segments. I wrote a bunch of code to do just that...and was floored when my partner pointed out to me that ffmpeg could give me that exact data on the commandline.

Any pointers? Would love to see that command

ffmpeg -i in.mp3 -af anequalizer=c0 f=70 w=100 g=-24 t=0|c1 f=70 w=100 g=-24 t=0, anequalizer=c0 f=4000 w=7500 g=-24 t=0|c1 f=4000 w=7500 g=-24 t=0 astats=metadata=1:reset=48,ametadata=print:key=lavfi.astats.Overall.RMS_level:file=log.txt -f null -

YAPE: yet another prores encoder

Sweet! Very excited to do quality benchmarks with this: https://ffmpeg.org/#aac_encoder_stable

I'm curious, how do you intend to benchmark it? Do you have some metric for audio quality or do you just mean blind listening tests?

There is PEAQ[1], so I suppose I should try implementations like GstPEAQ, Peaqb, and PQevalAudio. If anyone reading along has tried these and has thoughts on them or implementations I'm unaware of, I'd appreciate hearing about them.

Typically I just do informal listening tests, in this case against Apple's AAC encoder (the gold standard). You're right that ideally, I'd do something more formal like double-blind ABX testing.

[1] https://en.wikipedia.org/wiki/Perceptual_Evaluation_of_Audio...

quietly powering video literally everywhere. such an amazing project

I feel like I know Ffmpeg better than most, and I still only know between 5-10% of all it has to offer.

It’s amazing that they manage to keep adding features.

The swiss army knife of audio/video. I love ffmpeg

Reply on news.ycombinator.com