The AV1 Encoder Reality Check
AV1 has been "the future of video" since 2018. In 2026 it's finally usable in production, but only because the encoders matured. Three implementations dominate professional and prosumer work:
- SVT-AV1 (Intel): production-grade speed, decent quality
- libaom (Google): reference quality, painfully slow
- rav1e (Rust community): simple, low-memory, middle-ground speed
The three encoders produce AV1 files that any AV1 decoder plays. The compression quality and encoding speed are wildly different. Picking the wrong encoder for your workflow costs hours of compute or quality you can see.
This post compares the three at production-relevant settings, with specific recommendations for streaming, archival, and on-demand video. Our video compressor supports H.264 and HEVC for delivery; AV1 encoding workflows are upstream of that.
What AV1 Achieves
AV1 was designed to replace H.264 (delivery codec since 2003) and HEVC (royalty trap since 2013) with a royalty-free format. By the numbers in 2026:
| Codec | File size at same quality | Encoding speed | Decoding |
|---|---|---|---|
| H.264 (libx264) | Reference | Fast | Universal |
| HEVC (libx265) | 50% | Slow | Modern hardware |
| AV1 (SVT-AV1) | 35% | Reasonable | Modern hardware |
| AV1 (libaom) | 30% | Very slow | Modern hardware |
| AV1 (rav1e) | 35% | Moderate | Modern hardware |
AV1 saves real bandwidth at the same visual quality. The trade-off was always encoding speed. Now SVT-AV1 has narrowed that gap to "slower than H.264 but feasible for production."
SVT-AV1: The Production Default
Intel's Scalable Video Technology AV1 encoder is the practical choice for most workflows in 2026. Speed is roughly 1/3 of x264 for similar quality. Production settings:
ffmpeg -i input.mp4 \
-c:v libsvtav1 \
-preset 6 \
-crf 30 \
-g 240 \
-c:a libopus -b:a 128k \
output.av1.mp4
Reading the parameters:
preset 6: balance of speed and quality. Range 0-12, lower is slower/better.crf 30: quality target. Range 0-63, similar mental model to x264.g 240: GOP length (keyframe interval) of 240 frames = 8 seconds at 30fps.libopus 128k: AV1 pairs naturally with Opus audio.
For VOD (video on demand): preset 4-6, CRF 28-32. Slower preset = better quality, but encoding time blows up below preset 4.
For streaming (longer GOP, smoother bitrate): preset 8-10, larger GOP, CBR mode if needed.
For archival (highest quality regardless of speed): preset 2-4, CRF 18-22.
libaom: The Quality Reference
Google's libaom-av1 produces the best AV1 output at the cost of being the slowest. Production settings:
ffmpeg -i input.mp4 \
-c:v libaom-av1 \
-cpu-used 4 \
-crf 30 \
-b:v 0 \
-c:a libopus -b:a 128k \
output.av1.mp4
cpu-used 4 is the production sweet spot for libaom. Range 0-8, lower is slower/better. cpu-used 0 is reference quality but encoding can take days for a 4K source.
For libaom, the speed comparison is rough:
| cpu-used | Speed (relative to libx264) |
|---|---|
| 0 | ~1/100x |
| 2 | ~1/30x |
| 4 | ~1/10x |
| 6 | ~1/5x |
| 8 | ~1/3x |
cpu-used 4 produces output that beats SVT-AV1 preset 6 in detail retention, especially in dark scenes. The encoding time is roughly 5x longer.
Use libaom when quality matters more than speed: archival masters, premium VOD, scientific or legal evidence preservation.
rav1e: The Lightweight Option
Mozilla and Xiph's Rust-based encoder targets simplicity and low memory usage. Production settings:
ffmpeg -i input.mp4 \
-c:v librav1e \
-speed 6 \
-qp 100 \
-c:a libopus -b:a 128k \
output.av1.mp4
speed 6 is a balance. Range 0-10. Faster than libaom at the same cpu-used, slower than SVT-AV1 at the same preset.
rav1e shines on memory-constrained systems (embedded, mobile, low-RAM cloud instances). The encoder uses about 1/4 the RAM of SVT-AV1 at similar settings.
For most production: skip rav1e in favor of SVT-AV1 unless memory is the constraint.
Quality Comparison at Same Bitrate
For a 4K30 8-bit Rec.709 source at target ~5 Mbps:
| Encoder | Settings | Quality (VMAF) | Encode time |
|---|---|---|---|
| SVT-AV1 | preset 6 CRF 30 | 92.4 | 1.0x |
| SVT-AV1 | preset 4 CRF 30 | 93.1 | 2.5x |
| SVT-AV1 | preset 2 CRF 30 | 93.5 | 6.0x |
| libaom | cpu-used 4 CRF 30 | 93.2 | 4.5x |
| libaom | cpu-used 2 CRF 30 | 94.0 | 14x |
| rav1e | speed 6 qp 100 | 91.8 | 1.8x |
| libx265 | preset slow CRF 22 | 92.6 | 0.8x |
| libx264 | preset slow CRF 18 | 92.0 | 0.4x |
VMAF is a reference perceptual quality metric. Above 92 is "transparent" (most viewers don't perceive compression). Below 88 is "noticeable artifacts."
For most production: SVT-AV1 preset 6 is the right balance. libaom cpu-used 4 wins on quality if you can spare the time.
When AV1 Doesn't Make Sense
AV1 is the future, not the present, for some use cases:
Live streaming: AV1 encoding is too slow for 1080p60 real-time on most hardware. Hardware AV1 encoders (Intel ARC, NVIDIA RTX 40, AMD RDNA3) can do 1080p60 in real-time but at lower quality than CPU. For live: H.264 still wins.
Editing intermediates: AV1 decoding is computationally expensive. ProRes 422 or DNxHR is a better intermediate for editing. AV1 is for finished delivery.
Mobile delivery to old devices: iPhone 11 and earlier don't support AV1 hardware decode. Even iPhone 12-14 have inconsistent support. For broad mobile reach in 2026: H.264 first, HEVC second, AV1 third.
Email attachments: target file size for email is small enough that H.264 at high CRF works. AV1 saves bandwidth but adds the "what is this codec" question.
For practical AV1 use in 2026, the right targets are: YouTube delivery (re-encodes anyway), VOD platforms (Netflix, Disney+ both use AV1), and archival (long-term file size matters).
Hardware AV1 Encoding
Intel ARC GPUs, NVIDIA RTX 40 series, and AMD RDNA3 GPUs have hardware AV1 encoders:
# NVIDIA NVENC AV1 (RTX 40 series)
ffmpeg -i input.mp4 \
-c:v av1_nvenc \
-preset p7 \
-cq 30 \
-b:v 0 \
-c:a libopus -b:a 128k \
output.av1.mp4
# Intel QSV AV1 (ARC GPUs)
ffmpeg -i input.mp4 \
-c:v av1_qsv \
-preset slower \
-global_quality 30 \
-c:a libopus -b:a 128k \
output.av1.mp4
# AMD AMF AV1 (RDNA3)
ffmpeg -i input.mp4 \
-c:v av1_amf \
-quality quality \
-qp_p 30 \
-c:a libopus -b:a 128k \
output.av1.mp4
Hardware AV1 quality at the same bitrate is roughly equal to SVT-AV1 preset 8. Speed is 5-10x faster than CPU encoding. For batch work or live encoding, hardware AV1 is the right choice.
Pro Tip: RTX 40-series cards encode AV1 at near-realtime for 4K60. If you're producing AV1 deliverables at scale, the GPU upgrade pays for itself in compute time saved.
Container Choices for AV1
AV1 video can ship in several containers:
| Container | Compatibility | Use case |
|---|---|---|
| MP4 | Strong (since 2022) | Default for streaming and download |
| WebM | Strong | Google ecosystem, transparency support |
| MKV | Strong | Storage, multi-language tracks |
| ISOBMFF (mov) | Limited | Apple ecosystem |
For YouTube and most streaming: MP4. For Vimeo and OTT platforms: MP4 or WebM. For complex projects with multiple subtitle tracks: MKV.
For background on container choices, see Best Video Formats.
Audio Pairing With AV1
AV1's natural audio companion is Opus (royalty-free, designed alongside AV1):
| Audio codec | Bitrate (stereo) | Use case |
|---|---|---|
| Opus | 96-128 kbps | Optimal for AV1 video |
| AAC | 128-192 kbps | Compatibility with older players |
| FLAC | ~900 kbps | Lossless archival |
For social and streaming: Opus 128 kbps is the production default. For broader compatibility on platforms that don't yet handle Opus in MP4 (rare in 2026): AAC.
For background on audio bitrate, see Audio Bitrate Quality Guide.
Common Issues
Slow encoding: SVT-AV1 preset 6 on 4K is roughly real-time on a 16-core CPU. preset 2 is 6x slower. If encoding is too slow, increase preset number or use hardware encoding.
Banding in dark scenes: 8-bit AV1 with insufficient bitrate. Use 10-bit (-pix_fmt yuv420p10le) and higher bitrate (CRF 26 instead of 30).
Color shift on playback: incorrect color metadata. Add explicit primaries and transfer:
ffmpeg -i input.mp4 \
-c:v libsvtav1 -preset 6 -crf 30 \
-color_primaries bt709 -color_trc bt709 -colorspace bt709 \
-c:a libopus -b:a 128k \
output.av1.mp4
File won't play in older browsers: Chrome 90+, Firefox 83+, Safari 17+ support AV1. Older versions need a fallback. Use <picture> or <video> element with multiple <source> tags.
Frequently Asked Questions
Is AV1 worth the encoding time vs HEVC?
For YouTube and streaming: yes, the bandwidth savings compound across millions of views. For one-off shares: HEVC at the same quality encodes 5-10x faster, and the file size difference doesn't matter at small scale.
Can AV1 encoding use multiple GPUs?
SVT-AV1 is CPU-only and parallelizes well across cores. Hardware encoders use one GPU's encoder block. For batch work, run multiple files in parallel on separate GPUs.
What's the encoding speed gap between SVT-AV1 and libx264?
At quality-equivalent settings (SVT-AV1 preset 6 vs libx264 preset slow): SVT-AV1 is roughly 3-5x slower. Both can be parallelized across cores; SVT-AV1 scales slightly worse on very high core counts.
Should I worry about AV1 patent licensing?
AV1 is royalty-free under the AOMedia Patent License. Major patent holders (Google, Cisco, Mozilla, Intel, Microsoft, Netflix) cross-license freely. The MPEG LA-style royalty trap that HEVC fell into is not a concern for AV1.
Does AV1 support HDR?
Yes, AV1 supports HDR10 metadata and 10-bit color. Use -pix_fmt yuv420p10le and add color metadata flags. AV1 also supports Dolby Vision Profile 10 (rare in production).
How do I batch convert a folder to AV1?
for f in *.mp4; do
ffmpeg -i "$f" \
-c:v libsvtav1 -preset 6 -crf 30 \
-c:a libopus -b:a 128k \
"av1/${f%.*}.av1.mp4"
done
For batch conversion patterns, see Batch Processing Files Guide.
Related Reading
Bottom Line
For AV1 encoding in 2026: SVT-AV1 preset 6 CRF 30 is the production default. libaom cpu-used 4 wins on quality if encoding time isn't the constraint. Hardware AV1 (RTX 40, ARC, RDNA3) handles real-time and batch work cheaply. Pair with Opus 128k audio in MP4 container. Our video compressor handles the H.264/HEVC delivery side; AV1 ships through the same MP4 container.



