About AV2

AV2 is the next-generation video coding specification from the Alliance for Open Media (AOMedia). Building on the foundation of AV1, AV2 is engineered to provide superior compression efficiency, enabling high-quality video delivery at significantly lower bitrates. It is optimized for the evolving demands of streaming, broadcasting, and real-time video conferencing.

This specification serves as the definitive technical reference for AV2 implementations. It outlines the bitstream syntax, semantics, and decoding processes required to ensure full conformance.

AV2 provides enhanced support for AR/VR applications, split-screen delivery of multiple programs, improved handling of screen content, and an ability to operate over a wider visual quality range.

To assist implementers, the AOMedia Video Model (AVM) serves as the official reference software.

Please note that this is a draft release; ongoing refinements and known issues are detailed in the Release Notes and will be resolved before the final publication.

Notice

This specification is currently in draft status. We welcome feedback and issue reports via email at wg-codec-chair@aomedia.org, though please note that responses may be limited as this is a work-in-progress document.

Available Versions

Using the Specification

Full Specification

The complete AV2 coding specification document includes all sections from scope and definitions through annexes. It provides comprehensive coverage of the format, syntax, semantics, and decoding process.

Additional Tables

The Additional Tables are extracted lookup tables from section 9 as C header files, useful for implementation reference.

Syntax Browser

The syntax browser provides a specialized view of Sections 5 (Syntax Structures) and 6 (Semantics) in a split-pane interface. Features include:

• Side-by-side view of syntax definitions and their semantics
• Clickable syntax elements for easy navigation
• Search functionality across both sections
• Copy-to-clipboard for syntax structures

Reference Software

The reference software, known as AVM, corresponding to this version of the specification is research-v13.0.0 tag.

Release Notes

This document is intended to be used in conjunction with the released AV2 Draft Final Deliverable specification and corresponding AV2 Reference Software (AVM). Its purpose is to highlight known issues that will be addressed prior to the publication of the Final AV2 specification and reference software, and to make implementers aware that work on such issues is on-going. Issues are labeled as Spec and/or Software to indicate which aspect(s) of this release are impacted. A number of editorial improvements are planned for the final release to ensure correctness, and for the consistent use of language, naming, and formatting.

The AVM reference software, tagged "research-v13.0.0", is available at: https://gitlab.com/AOMediaCodec/avm/-/tree/research-v13.0.0?ref_type=tags

As reference software, AVM does not support every conceivable use case, and it has not been fully optimized for production deployment.

Download Release Notes as PDF

Terminology

Draft Specification draft of "AV2 specification v1.0.0".

Reference Software reference software (AVM research-v13.0.0) for the Draft Specification.

Coding tools

1. Spec, softwareRestricted switch frames: The current specification does not describe how Quantization Matrices and Film Grain modeling are expected to be handled when Restricted Switch Frame OBUs are encountered in a bitstream.
2. Spec, softwareCoding block sizes: There is a known issue in the current specification where content using the YUV 4:2:2 chroma format is allowed to use a 4x64 block size. This will be corrected in the final specification, such that 4x64 and 64x4 coding block sizes are disallowed in all chroma formats.
3. SoftwareQuantization matrices: There is a known issue in the Draft Software where the subsampling process used to generate 4x32 and 32x4 quantization matrices does not match the Draft Specification, which follows the AV1 subsampling scheme. The final version of the software will be corrected to match the Draft Specification. Multi-layer Support

AV2 will support advanced multilayer features and capabilities that are currently not fully described in the Draft Specification or exercised in the Reference Software. The following aspects will be further clarified in the final version of the specification and reference code:

1. SpecSemantics for several multilayer related features, such as the semantics of the OPS OBU, are not well defined or incorrect or missing.

2. SpecThe current specification does not clearly describe how different layers, carrying data annotated with different types of information, such as view IDs, auxiliary information, atlas segments etc., can be associated and used for different applications (e.g. in the context of a stereoscopic application that includes transparency or other types of information such as depth). This will be clarified in the final specification. Examples for applications such as sub-region coding using layers, will also be included.

3. Spec, softwareSub-bitstream extraction based on operating point selection is missing from both the Draft Specification and the Draft Software. Temporal units and random access support

4. SpecThe definitions of a temporal unit and the random access points need refinement. In particular, the random access point definition is not complete. Additional requirements beyond the presence of certain OBU types (e.g. a CLK or OLK OBU) are necessary. For example, the availability of any corresponding HLS information (such as an SH OBU) either within the same temporal unit or through external means is also essential for such definitions. More specifically, the definition of the following random access points is missing:

• Closed Loop Key Frame Random Access
• Open Loop Key Frame Random Access
• Random Access Switch Frame

1. SpecActivation of certain HLS OBUs, such as SH, LCR, and MFH OBUs, is not clearly defined. Profile/Level Capabilities

2. SpecCompatibility of bitstreams tagged with different "Sequence" level profile indication syntax elements is currently not specified in the Draft Specification. For example, bitstreams tagged with seq_profile_idc = 0, can be decoded by any decoders that are capable of Main_420_10 decoding. However, bitstreams with 0 < seq_profile_idc < 4, chroma_format_idc = 0, and seq_max_mlayer_cnt = 1 are also decodable by a Main_420_10_IP0 decoder. This needs to be more clearly described in the specification with a well defined profile "compatibility" section.

3. Spec, softwareThe Draft Specification and Draft Software do not properly address the fact that only profiles capable of 8 and 10 bit support are defined. The Draft Specification and the Draft Software assume that there is support for up to 12 bits, and incorrectly set variables for bit_depth_idc > 1 to states relevant to 12 bit decoding. This needs to be addressed. In particular, The Draft Specification should enforce that bit_depth_idc shall not take values greater than 1, and that other values are reserved for future versions of this specification. The related sections of the specification and / or code modules of the reference SW may need to be adjusted.

4. Spec, softwareThe Draft Specification does not include conformance constraints for the presence of the MSDO and LCR OBUs for different multilayer bitstream configurations and profiles.

5. Spec, softwareSome level related constraints on frame buffer management when both Global IBC and in-loop filtering are enabled are not correctly reflected in the Draft Specification. Such constraints result in allocating one frame slot in the decoder buffer pool to hold the reconstructed frame before the in-loop filter is applied.

6. Spec, softwareThe 4:2:2 and 4:4:4 profiles will be indicated with separate seq_profile_idc values. Metadata

7. SpecIt might be appropriate to move the specific metadata payload types into a separate section (e.g. an Annex) and not include them in sections 5 and 6 with essential decoding information. Such a section should also describe how metadata persistence and inheritance should be handled, which is currently missing.

8. Spec, softwareInheritance of information across embedded layers for Content Interpretation (CI) OBUs is currently not described

9. Spec, softwareIssues were identified with the Buffer Removal Timing Metadata.

10. SpecThe syntax of the ITU-T T.35 metadata is incomplete since it does not clearly identify the payload as syntax (it is identified as such in a note, which is not appropriate).

11. Spec, softwareThe temporal point info metadata is planned to be changed in the final version of the specification.

12. Spec, softwareNew informative metadata types may be added in the final version of the specification. Decoder model

13. Spec, softwareThere is ongoing discussion whether decoder model conformance needs to be required to simultaneously account for both models specified in the specification rather than just one model. This may result in changes in the final version of the specification and the reference software.

14. Spec, softwareDue to the changes in random access and layer handling in the AV2 design compared to that of AV1, as well as how the output process is handled, additional changes to the details of the decoder model may need to be made in the final version of the specification. Other items

15. Spec, softwareThe AV2 specification defines numerous syntax elements and variables that are associated with elements in the CICP specification (ISO/IEC 23091-2 and ITU-T H.273). These references are sometimes incorrect, i.e. use of ISO/IEC 23091-4 throughout, which is only a technical report. In addition the CICP related tables should likely be removed or edited and instead the CICP tables should be directly referenced. This can ensure that if any new values are introduced in the CICP specification, those will be immediately supported without requiring publication of a new version of the specification. Edited tables that provide only samples of values that are useful could still be retained.

16. Spec, softwareThere are known differences between the Draft Specification and the Draft Software related to syntax elements show_frame and showable_frame. These changes will be aligned and further clarified in the final specification and reference software. Such changes will improve readability and the understanding of how such features can be used and/or implemented.

17. Spec, softwareOBU extensibility is not properly reflected in the Draft Specification and will be addressed in the final version of the specification.