3 ms·

We went through implementing AVIF support in our image service, only to find out that it generated larger images than WebP anyway. Not sure if that’s typical th

by rmccue 6y ago

We went through implementing AVIF support in our image service, only to find out that it generated larger images than WebP anyway. Not sure if that’s typical though.

lucb1e 6y ago
Do you have an example? The F1 image from the post above looks pretty convincing, if you have different data then I'd be curious to see that. For example you say larger images, but what about quality? Was that not better for the larger file size? Part of this is subjective as well (I really mind the non-smooth areas and text artifacts in jpeg, other people might mind other things).
- astrange 6y ago
  It's actually pretty easy to fix those artifacts in a JPEG decoder at the cost of making it a little too smooth. The text artifacts are only partially a compression artifact and partially from a poor quality YUV>RGB conversion, so you can do a better one. Blocking artifacts can be reduced with a deblocking filter; JPEG/MPEG1/MPEG2 are so simple it doesn't take much work to guess what was supposed to be there.
efortis 6y ago
If you are using `avifenc` try with `--speed 0`. If you are doing that, or using `colorist` then tune the quantizer. After experimeting with the quantizer, these are the settings I use for screenshots with text and sharp lines: https://blog.uidrafter.com/engineering/convert-to-avif-programmatically https://blog.uidrafter.com/engineering/convert-to-avif-progr...
giorgioz 6y ago
Yes we also try converting images to .avif with the NPM module Sharp version 0.27.0 and we images as large as .webp using the default speed of 5/8 and it takes 10 seconds to encode on the same AWS Lambda of 512mb. I red that increasing the speed to 8/8 will lead to smaller file but the encoding time will take 3 minutes. https://fronius.me/articles/2020-10-14-comparing-image-formats-jpg-webp-avif.html#:~:text=Processing%20speed&text=When%20I%20started%20out%20my,and%20libvips%20(as%20of%208.10 https://fronius.me/articles/2020-10-14-comparing-image-forma...

by rmccue 6y ago

We went through implementing AVIF support in our image service, only to find out that it generated larger images than WebP anyway. Not sure if that’s typical though.

lucb1e 6y ago

Do you have an example? The F1 image from the post above looks pretty convincing, if you have different data then I'd be curious to see that. For example you say larger images, but what about quality? Was that not better for the larger file size? Part of this is subjective as well (I really mind the non-smooth areas and text artifacts in jpeg, other people might mind other things).

astrange 6y ago
It's actually pretty easy to fix those artifacts in a JPEG decoder at the cost of making it a little too smooth. The text artifacts are only partially a compression artifact and partially from a poor quality YUV>RGB conversion, so you can do a better one. Blocking artifacts can be reduced with a deblocking filter; JPEG/MPEG1/MPEG2 are so simple it doesn't take much work to guess what was supposed to be there.

efortis 6y ago

If you are using `avifenc` try with `--speed 0`. If you are doing that, or using `colorist` then tune the quantizer. After experimeting with the quantizer, these are the settings I use for screenshots with text and sharp lines: https://blog.uidrafter.com/engineering/convert-to-avif-programmatically https://blog.uidrafter.com/engineering/convert-to-avif-progr...

giorgioz 6y ago

Yes we also try converting images to .avif with the NPM module Sharp version 0.27.0 and we images as large as .webp using the default speed of 5/8 and it takes 10 seconds to encode on the same AWS Lambda of 512mb. I red that increasing the speed to 8/8 will lead to smaller file but the encoding time will take 3 minutes. https://fronius.me/articles/2020-10-14-comparing-image-formats-jpg-webp-avif.html#:~:text=Processing%20speed&text=When%20I%20started%20out%20my,and%20libvips%20(as%20of%208.10 https://fronius.me/articles/2020-10-14-comparing-image-forma...