My main use case is modifying youtube videos of tech tutorials where the speaker overlays a video of themselves in a corner of the video. drawvg is used to blackout that area of the video. I'm sure some viewers like having a visible talking head shown on the same screen as the code but I find the constant motion of someone's lips moving and eyes blinking in my peripheral vision extremely distracting. Our vision is very tuned into paying attention to faces so the brain constantly fighting that urge so it can concentrate on the code. (A low-tech solution is to just put a yellow sticky know on the monitor to cover up the speaker but that means you can't easily resize/move the window playing the video ... so ffmpeg to the rescue.)
If the overlay was a rectangle, you can use the older drawbox filter and don't need drawvg. However, some content creaters use circles and that's where drawvg works better. Instead of creating a separate .vgs file, I just use the inline syntax like this:
ffmpeg -i input.webm -filter_complex "[0:v]drawvg='circle 3388 1670 400 setcolor black fill'[v2];[0:a]atempo=1.5[a2]" -map "[v2]" -map "[a2]" output.mp4
That puts a black filled circle on the bottom right corner of a 4k vid to cover up the speaker. Different vids from different creators will require different x,y,radius coordinates.(The author of the drawvg code in the git log appears to be the same as the author of this thread's article.)
[1] https://git.ffmpeg.org/gitweb/ffmpeg.git/commit/016d767c8e9d...
On the other hand, when I compared the binaries (ffmpeg, ffprobe, ffplay) I downloaded the other day with the ones I had installed since around September, they where almost 100 MB larger. I don't remember the exact size of the old ones but the new ones are 640 MB, the old ones well under 600 MB. The only difference in included libraries was Cairo and the JPEG-XS lib. So while I think a bunch of new ML models would be really cool, maybe they don't want to go down that route. But some kind of pluggable system with accelerated ML models would be helpful I think.
Is there something similar that supports shaders? Like metal / wgsl / glsl or something?
Sounds like a fun project...
Still, I find the syntax it uses horrible:
ffmpeg -an -ss 12 -t 3 -i bigbuckbunny.mov -vf 'crop=iw-1, drawvg=file=progress.vgs, format=yuv420p' -c:v libvpx-vp9 output.webm
I understand that most of this comes from simplicity of use from the
shell, so if you take this point of view, the above makes a lot of sense.My poor, feeble brain, though, has a hard time deducing all of this. Yes, I can kind of know what it does to some extent ... start at 12 seconds right? during 3 seconds ... apply the specified filter in the specified format, use libvpx-vp9 as the video codec ... but the above example is somewhat simple. There are total monsters in actual use when it comes to the filter subsystem in ffmpeg. Avisynth was fairly easy on my brain; ffmpeg does not, and nobody among the ffmpeg dev team seems to think that complicated uses are an issue. I even wrote a small ruby script that expands shortcut options as above, into the corresponding long names, simply because the long names are a bit easier to remember. Even that fails when it comes to complex filters used.
It's a shame because ffmpeg is otherwise really great.