H.264 vs AV1: Equal-Quality Measurement
At interpolated SSIM 0.992 on the measured clip, AV1 used 3,402.6 kbps and H.264 used 3,925.9 kbps, so this AV1 encode was about 13.3% smaller. This is a result for a Big Buck Bunny excerpt and the stated libsvtav1 and libx264 presets, not a universal AV1 saving. Use H.264 for the broadest documented upload compatibility; use AV1 only for a path that explicitly accepts or serves it.
Short answer
Choose H.264 for the broadest upload acceptance and AV1 only where the path explicitly accepts it. At matched SSIM 0.992 on our test clip, AV1 used 3,402.6 kbps against 3,925.9 kbps for H.264, about 13.3% smaller.
Measured data
The video run used a 10 second, 1920 x 1080, 30 fps excerpt of Big Buck Bunny, an animated film rather than camera footage, licensed CC BY 3.0 by the Blender Foundation. Results describe this clip and these encoder settings, not every video or any platform pipeline.
| Codec and encoder | Measured SSIM | Interpolated bitrate | Preset |
|---|---|---|---|
| H.264, libx264 | 0.992 | 3,925.9 kbps | medium |
| AV1, libsvtav1 | 0.992 | 3,402.6 kbps | 8 |
The approximate equal-quality value is a linear interpolation between two bracketing CRF rungs, not a direct encode at exactly SSIM 0.992.
Arithmetic: (1 - 3,402.6 / 3,925.9) x 100 = 13.3% smaller on this run.
When to use each option
Use H.264 for delivery certainty
Choose it when the platform's own page recommends H.264 or does not publish AV1 upload acceptance.
Use AV1 where support is explicit
AV1 can reduce bitrate, but upload acceptance, playback decode and encoder cost are separate decisions.
Do not use the percentage as a promise
Animation, camera footage, grain and motion produce different curves. Measure a representative source before planning storage.
Questions people ask
Was AV1 13.3% smaller in every test?
No. It was 13.3% smaller at the approximate matched SSIM point on this one animated excerpt.
Why not compare CRF 24 with CRF 24?
The scales do not represent the same quality across encoders, so equal CRF would answer the wrong question.
Does YouTube playback in AV1 prove AV1 uploads are accepted?
No. Playback delivery and upload acceptance are different claims. Use the official upload page for upload guidance.
Is SSIM a pass score?
No. It is a full-frame similarity metric. It helps match this run but does not define subjective quality for every viewer.
Measured vs sourced
Cells marked measured are Koldflux Media Lab measurements. Cells marked sourced derive from each platform / codec standards body. Cells marked guidance are expert guidance, not measured.
Sources
- FFmpeg codec documentation (read 2026-10-08)
- Alliance for Open Media: AV1 features (read 2026-10-08)
- Blender Foundation: Big Buck Bunny and CC BY 3.0 (read 2026-10-08)
Verified at 2026-10-08.
Next steps
Inspect my file: /tools/validate
Convert / make correct version: /tools/convert
Create all target versions: /tools/pack