Missing from the article is rav1e, which encodes AV1, and hence AVIF, a lot faster than the reference implementation aom. I've had cases where aom would not finish converting an image in a minute of waiting what rav1e would do in less than 10 seconds.
rav1e is generally head to head with libaom on static images, and which one wins on the speed/quality/size frontier depends a lot on the image and settings, as much as +/- 20%. I suspect rav1e has an inefficient block size selection algorithm, so the particular shape of blocks is a make or break for it.
I’ve only compared rav1e to mozjpeg and libwebp, and at fastest speeds it’s only barely ahead.
