AO Media Says ‘The Future of Innovation is Open’
A perennial interest at the 8K Association is the topic of codecs and so we were interested to see a webinar produced by AOMedia, the organisation that is developing the AV1 and AV2 codecs. The full webinar available online at the time of this publication. The title of the event was ‘The Future of Innovation is Open’.
The webinar was moderated by Dr. Pierre-Anthony Lemieux, AOMedia Executive Director. (director@aomedia.org). Lemieux first introduced the organisation which has strong membership from technical companies in the content, hardware, software and services fields. Lemieux said that it has a ‘typical standards organization structure’. Specifications are developed under a royalty-free patent policy and are made publicly available on the website. That’s a key point of the AV1/AV2 technologies, in comparison to the MPEG-derived codecs such as HEVC and VVC.
Lemieux introduced the standards including the AV1, an Emmy award winning codec. AVIF is a image data format based on AV1 and IAMF is for immersive audio. SVT-AV1 is an optimized encoder for AV1 that won an Innovation award at IBC. The group is working on media encoding for volumetric video (Static Polygonal Mesh Coding) and also on AV2, the next codec and which had been due at the end of 2025 (the webinar took place just before the end of the year).
The participants on the webinar were
- Debargha Mukherjee – Principal Engineer (Google)
- Anush Moorthy – Engineering Leader – Video Algorithms & Media Systems (Netflix)
- Adrian Grange – Senior Technical Program Manager (Google)
- Hassene Tmar – Technical Program Manager (Meta)
Mukherjee started with a presentation that outlined AV2 – he has acted as a leader of the development. His first chart was a timeline of codec developments from 1990 to 2030.
As much as 75% or 80% of internet traffic is now video. VP9 was the first open standard from WebM and then AOMedia was formed. AV1 was released in 2018 and from 2020 work started in 2020 on AV2. AV1 was adopted by YouTube, Meta and other streamers. The requirements for AV2 suggested a need for 40% lower bandwidth than AV1 and with a PSNR target of 30% lower. Decoder complexity and silicon area was targeted at no more than double that of AV1.
The first development phase was on technical topics such as the choice of test sequences and conditions, as well as the processes for standardization. The AOM Video Model (AVM) was established. Many tools were developed subsquently and the AVM was updated every four months with perceptual tests after each of the last few releases. Seventeen technology companies were involved in providing tools or giving feedback on technology.
Not an AI Codec
Mukherjee pointed out that AV2 is a ‘conventional hybrid’ codec, not an AI-based codec although he sees it as a step towards AI codecs as it has a number of data-driven tools. He listed some of the detailed technology improvements.
Features include a hardware efficient design; very good efficiency for screen graphics rather than just ‘natural’ content; multi-layer and multi-stream features are included for mosaic and immersive video and better scalability for the future; lossless encoding and film grain synthesis are significantly improved and there should be support for 4:2:2 and 4:4:4 video as well as 4:2:0.
Low level coding tools have been finalized and the group is at the last stages of higher level coding. Mukherjee said that the random access case is arguably the most important use case and here AV2 is showing 28% to 32% improvement in VMAF and PSNR over AV1. That’s the same level of improvement as AV1 was, over VP9. He also showed the table of full results of perceptual testing which showed almost 40% improvement – matching the target.
Netflix Looks at AV2 Features & AV1 Deployment
Moorthy of Netflix then gave a presentation before the full panel session. He summarized the rollout plan which will be through standardization, ecosystem support and then to productization. Standardization also includes the development of unique tools to give the codec a competitive advantage against other technologies.
Ecosystem support is mostly about ensuring that there are software and hardware decoders available, as well as encoders available from vendors. Productization will happen once these factors are in place and Moorthy highlighted that once you use the technology at scale, you learn things about it. He explained FGS which we have previously reported on in some detail.
Other unique features of AV2 are S-frames that improve stream switching and error recovery. Reference Picture Resampling allows in-GOP resolution changes without needing new key-frames. Enhanced screen coding and scalable coding tools are available in the main profile to provide wider decoder support.
Turning to the question of the ecosystem, dav1d is an open-source, cross-platform software decoder that provides great performance even on low end hardware, Moorthy said. It is the default AV1 codec on Chrome, Firefox and is available for Android. Hardware support of AV1 decoding is now almost ‘ubiquitous’ now with 88% of large screen hardware devices submitted for Netflix certification having hardware decoding. In the last two years, this has moved close to 100% and from 2025, support is mandatory for Netflix devices.
In the mobile sphere, high-tier Android and all iPhones since iPhone 15 Pro/iPhone 16 have hardware support for AV1, with the mid-tier increasing in proliferation. In the low-end devices, there is less support, but software decoding has been demonstrated by Meta, without excessive battery drain.
There are multiple hardware- and software-based encoding vendors including Allegro, Ateme, Bitmovin and Chips&Media among others for different use cases.
Moorthy said that there can be a misunderstanding about power consumption as AV1 is an advanced codec. It may be thought that a software encoder, in particular may mean a lot of power consumption, but this is not the case according to work by Meta/Youtube and Vodafone.
There have been a number of success stories in deployment. 70%+ of Meta’s global media watch time uses AV1 and the firm uses SVT-AV1 fast decode to reduce decoder complexity by 25%. AV1 us used for video calls on WhatsApp, Messenger and Instagram Direct. Agora also has a PaaS (Platform as a Service) product that Moorthy claimed has a complexity of just 75% of the x264 version, while achieving a 42% reduction in bitrate. Over a third of YouTube watched is now in AV1 while 75% of the watch-time weighted catalog is in AV1. It is also being used in popular live streams.
Netflix has all of its SDR VoD titles available in AV1 and it is the second most streamed profile, below HEVC, but well above AVC). The company expects AV1 to hit the top in 2026. Microsoft has been using AV1 since 2023 for Teams and screen sharing with ‘sharper text and smoother scrolling’ at 63% lower bitrates than H.264.
Moorthy said that the reasons for the deployment are:
- An increase in quality (Netflix saw 10pt VMAF boost and 5% increase in 4K streaming)
- A reduction in bitrate (Meta saw 48%-65% reduction in bits at equal quality, Netflix got 31% reduction from FGS)
- Quality of Experience improves (Netflix saw 2-10% reduction in play delay and reduced playback errors as well as a 10% reduction in rebuffers using FGS)
This concluded the first part of the webinar and next week, we’ll bring the second part which was a group discussion on the codecs. We’ve also reached out to AO Media to ask why the AV2 release has not happened although it was widely flagged (including on the webinar) as being due at ‘the end of 2025’.
The second part of our report has now been published.
