Class-Leading Performance
We've tuned our AVIF performance via industry-standard benchmarks for perceptual fidelity including SSIMULACRA2 (graphs below). However, we did not just stop at metric scores; there are always outliers in a metric's correlation with our vision, so SVT-AV1-PSY's AVIF results have undergone extensive visual inspection and additional tuning to ensure consistency is preserved in any source.
Visual Comparisons
Visual comparisons between codecs are difficult. Images can be cherry-picked to flaunt a particular encoder's strengths, visual preferences differ across human beings, and metrics often don't tell a complete story when used as the basis for making fidelity matched comparisons. You have been warned that this section is rather unscientific, but we've tried our best to create fair comparisons here in the low to medium fidelity range to make differences easily visually distinguishable.
⚠️ JPEG XL: If you aren't using a browser that supports JPEG XL, the JXL images in these comparisons won't display. We would recommend using a browser supporting JXL for viewing these comparisons adequately; some examples are Thorium, Waterfox, & Safari.
Size Matched
Fidelity Matched
Methodology
If you want to dive deeper into our testing, see the methodology page.