Add support for AVIF #130
Add support for AVIF #130
Conversation
This will run the test-filesizes in travis on each build. It also compares the baseline file sizes in `fixtures.json`, so if we ever introduce chances that lead to large file sizes (or smaller) we'll get that information in the travis build (files growing larger than the fixture will also fail travis builds).
Want to make sure travis detects the failure
Sharp should soon be releasing AVIF support. This PR uses the alpha-3 version of libvips and upcoming version of Sharp to add support for avif images. This should work the same as webp: there is an X-Avif header or `avif=1` param, and the detection is expected to be handled by a lamnda function the same as WebP (check for `accept:image/avif` header). The results in `test` are good and bad: ``` ┌───────────────┬───────────────┬────────────────────┬──────────┬──────────┬──────────┬───────────────┬───────────────┐ │ Image │ Original Size │ Tachyon Size │ 100px │ 300px │ 700px │ 700px webp │ 700px avif │ ├───────────────┼───────────────┼────────────────────┼──────────┼──────────┼──────────┼───────────────┼───────────────┤ │ briefing-cop… │ 442.52 KB │ 112.67 KB (25%) │ 3.01 KB │ 9.34 KB │ 28.51 KB │ 15.31 KB (53… │ 14.57 KB (51… │ │ hdr.jpg │ 222.58 KB │ 145.56 KB (65%) │ 10.4 KB │ 23.53 KB │ 85.59 KB │ 80.86 KB (94… │ 111.09 KB (1… │ │ icons.png │ 94.24 KB │ 27.37 KB (29%) │ 3.89 KB │ 11.31 KB │ 26.94 KB │ 23.93 KB (88… │ 13.25 KB (49… │ │ Website.png │ 32.38 KB │ 33.78 KB (104%) │ 3.26 KB │ 13.27 KB │ 33.78 KB │ 19.81 KB (58… │ 22.21 KB (65… │ │ humans.png │ 2.22 MB │ 853.21 KB (37%) │ 8.96 KB │ 54.66 KB │ 273.89 … │ 139.19 KB (5… │ 174.48 KB (6… │ └───────────────┴───────────────┴────────────────────┴──────────┴──────────┴──────────┴───────────────┴───────────────┘ ``` In some cases it's less than webp, and in some cases more. I think it might be to do with these comparisons being quality:100, where we might do better at slightly dropping quality.
|
Oh nice. Have the security issues upstream in libvips been fixed? |
|
Apparently so! Release imminent
… On Dec 19, 2020, at 15:05, roborourke ***@***.***> wrote:
Oh nice. Have the security issues upstream in libvips been fixed?
—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub, or unsubscribe.
|
|
I think we might want to default webp over avif still, I included a script to compare real-world data in webp images, creating AVIF versions of the same, and comparing on average. I compared 1000 popular image requests to Tachyon, which resulting in the following: |
|
FYI based off https://www.industrialempathy.com/posts/avif-webp-quality-settings/, I applied the |
|
Also with new test-filesize: |
|
@joehoyle Is there a plan to continue this implementation? |
|
@johnbillion it's not scheduled, but yes given an infinite time scale, we'll want to do this. The pressure to do it reduced somewhat from the findings though, as it's not a lot smaller than webp, and now iOS support webp too then the coverage and gains from webp are still substantial. I think it would still be good to add though as per the above 9% improvement in file sizes. The other side to that is avif is about 3x slower to generate. That might have changed if the avif encoder has been more optimized since last year though. |
|
Sharp is at version 0.30.7 now (0.29.3 in this PR). Could be worth upgrading. |
|
Cool, I think we could pick this up after #162 is merged. We could still add AVIF as opt-in. |
Sharp should soon be releasing AVIF support. This PR uses the alpha-3 version of libvips and upcoming version of Sharp to add support for avif images.
This should work the same as webp: there is an X-Avif header or
avif=1param, and the detection is expected to be handled by a lamnda function the same as WebP (check foraccept:image/avifheader).The results in
testare good and bad:In some cases it's less than webp, and in some cases more. I think it might be to do with these comparisons being quality:100, where we might do better at slightly dropping quality