I was going to upload a >100 MiB video, but E621 has a 100 MiB limit for video posts. I tried lowering the quality settings, but that leaves highly visible compression artifacts. The encoder is "libvpx-vp9" and the speed is set to 0 .

You have to.

I actually haven't played around with -speed option myself. I think that's actually legacy command for -cpu-used and I think the default is already 0, meaning highest quality and longest time.

What actually effects efficiency of the encode is -deadline (legacy -quality), where options are realtime, good, best and placebo where default is good. best does increase encoding time immensily, but will make certain that every bit is squeezed, so with long videos needing to fit 100 MB I'll leave machine to crunch over night or in the background.

Another thing that helps is two pass, VP9 has many features which are not utilized without using that and also makes constrained quality much more accurate to use to get as near 100 MB as possible.

https://trac.ffmpeg.org/wiki/Encode/VP9

If you are in situation where the video gets way too compressed to hit the filesize limit, then downscale is sometimes better option, after all pixels do not matter if it's turning into a blur. Usually this applies if talking of HD content to begin with and it's usually good idea to go half dimensions directly (e.g. 2160p->1080p, 1080p->540p, etc.) with lanczos to avoid scaling effecting quality. Downscaling already small content (=<480p) even smaller will only make things even worse.

Mairo said:
I actually haven't played around with -speed option myself. I think that's actually legacy command for -cpu-used and I think the default is already 0, meaning highest quality and longest time.

What actually effects efficiency of the encode is -deadline (legacy -quality), where options are realtime, good, best and placebo where default is good. best does increase encoding time immensily, but will make certain that every bit is squeezed, so with long videos needing to fit 100 MB I'll leave machine to crunch over night or in the background.

Another thing that helps is two pass, VP9 has many features which are not utilized without using that and also makes constrained quality much more accurate to use to get as near 100 MB as possible.

https://trac.ffmpeg.org/wiki/Encode/VP9

If you are in situation where the video gets way too compressed to hit the filesize limit, then downscale is sometimes better option, after all pixels do not matter if it's turning into a blur. Usually this applies if talking of HD content to begin with and it's usually good idea to go half dimensions directly (e.g. 2160p->1080p, 1080p->540p, etc.) with lanczos to avoid scaling effecting quality. Downscaling already small content (=<480p) even smaller will only make things even worse.

I would not leave my computer unsupervised because i fear it could start a fire. I think it would be better to transcode in the background at weekend. Thanks by the way.

Mairo said:
I actually haven't played around with -speed option myself. I think that's actually legacy command for -cpu-used and I think the default is already 0, meaning highest quality and longest time.

What actually effects efficiency of the encode is -deadline (legacy -quality), where options are realtime, good, best and placebo where default is good. best does increase encoding time immensily, but will make certain that every bit is squeezed, so with long videos needing to fit 100 MB I'll leave machine to crunch over night or in the background.

Another thing that helps is two pass, VP9 has many features which are not utilized without using that and also makes constrained quality much more accurate to use to get as near 100 MB as possible.

https://trac.ffmpeg.org/wiki/Encode/VP9

If you are in situation where the video gets way too compressed to hit the filesize limit, then downscale is sometimes better option, after all pixels do not matter if it's turning into a blur. Usually this applies if talking of HD content to begin with and it's usually good idea to go half dimensions directly (e.g. 2160p->1080p, 1080p->540p, etc.) with lanczos to avoid scaling effecting quality. Downscaling already small content (=<480p) even smaller will only make things even worse.

I understood everything you said, i think i am an A/V nerd.