AVIF: brands and items

An AVIF file is a HEIF file, and also a MIAF file, whose images are coded with AV1. The box layout, item information, locations, references and grids are exactly those of HEIF, described on the HEIC pages: Boxes and items and Derived images and metadata. This page covers what AVIF adds, from the AVIF specification version 1.2.0 and the AV1 ISOBMFF binding; its properties are on Item properties.

Brands

Brand	Meaning
`avif`	Image or collection: the primary item is an AV1 image item, or a derived image (such as a grid) built from AV1 image items.
`avis`	Image sequence: the file holds one or more AV1 image sequences, and may hold auxiliary sequences.
`avio`	Intra-only: the primary item and every item it uses are made only of intra frames; with sequences, at least one sequence is all sync samples.
`MA1B`	The AVIF Baseline profile.
`MA1A`	The AVIF Advanced profile.

A still image should list avif, mif1, miaf and a profile brand; a sequence avis, msf1, miaf and a profile brand. With avif or avis as the major brand the extension should be .avif and the media type image/avif. A file might start 00 00 00 1C 66 74 79 70 61 76 69 66 00 00 00 00 61 76 69 66 6D 69 66 31 6D 69 61 66: ftyp, major brand avif, minor version 0, compatible avif, mif1, miaf.

Profiles

Profile	Images	Sequences	Largest coded image
Baseline (`MA1B`)	AV1 Main profile, level 5.1 or lower	Main profile, level 5.1 or lower	8,912,896 pixels, at most 8192 wide and 4352 high
Advanced (`MA1A`)	AV1 High profile, level 6.0 or lower	Main or High profile, level 5.1 or lower	35,651,584 pixels, at most 16384 wide and 8704 high

The pixel limits are informative and come from the levels. Larger pictures are stored as a grid of smaller AV1 items. AV1 itself carries 8, 10 or 12 bits per sample; deeper images use a Sample Transform derived item (below).

av01: the AV1 image item

Item type av01. It must have exactly one av1C property, which should be marked essential.
Its data is exactly what an AV1 sync sample would be: one temporal unit holding exactly one Sequence Header OBU and the frame. Setting still_picture and reduced_still_picture_header to 1 is recommended where possible.
Each OBU carries its size (obu_has_size_field 1), except that the last may leave it out and fill the rest of the data.
Tile list OBUs are not allowed; temporal delimiters, padding and redundant frame headers should not be used.
ispe gives the upscaled width and frame height of the output frame.

OBU header

From the AV1 bitstream specification, section 5.3.

Bits	Name	Description
1	obu_forbidden_bit	0.
4	obu_type	From the table below.
1	obu_extension_flag	1 when an extension byte follows.
1	obu_has_size_field	1 when a size follows.
1	obu_reserved_1bit	0.
3 + 2 + 3	temporal_id, spatial_id, reserved	The extension byte, when present: which temporal and spatial layer the OBU belongs to.
leb128	obu_size	Payload bytes, as little-endian base-128: 7 bits per byte, high bit set on every byte but the last.

obu_type	OBU
1	Sequence header
2	Temporal delimiter
3	Frame header
4	Tile group
5	Metadata
6	Frame (frame header and tile group together)
7	Redundant frame header
8	Tile list
15	Padding

Types 0 and 9 to 14 are reserved. A sequence header OBU usually starts 0A: type 1, size field present.

Alpha and depth

Alpha is an auxiliary AV1 item with an auxl reference to the colour item and auxC type urn:mpeg:mpegB:cicp:systems:auxiliary:alpha. Its sequence header has mono_chrome 1 and color_range 1 (full range), and its bit depth matches the colour item. It should have no colr, and readers ignore one if present.
A prem reference from the colour item to the alpha item says the colour is premultiplied.
Depth uses type urn:mpeg:mpegB:cicp:systems:auxiliary:depth, with the same constraints.
Animated alpha is an auxiliary track alongside the colour track.

Image sequences

An animated AVIF holds AV1 tracks as in MP4: a moov box whose track handler is pict, with exactly one av01 sample entry holding av1C. Each sample is one temporal unit. Several sequence header OBUs are allowed only if they are identical. A still primary item is usually stored as well, for readers that do not play sequences.

Grids, gain maps and sample transforms

Grids work as in HEIF. Transformative properties (clap, irot, imir) may sit only on the grid item, not on the tiles it is built from.
Gain maps: a tmap derived item (defined in ISO 21496-1) combines a base image with a gain map image to give an alternative rendition, usually HDR. The base image and the tmap item should be in an altr group, and the gain map item should be hidden.
Sample transform (sato, version 0) derives an image from others by an expression on their samples, for bit depths above 12. Its bit_depth field, 0 to 3, selects 8-, 16-, 32- or 64-bit arithmetic; expression tokens 0 are constants, 1 to 32 input images, 64 to 67 unary operators and 128 to 137 binary operators. The full operator list is in the AVIF specification and is not repeated here.

.AVIF File Format – Brands and items