AV1 Image File Format (AVIF)

1. Scope

[AV1] defines the syntax and semantics of an AV1 bitstream . The AV1 Image File Format ( AVIF ) defined in this document supports the storage of a subset of the syntax and semantics of an AV1 bitstream in a [HEIF] file. The AV1 Image File Format defines multiple profiles, which restrict the allowed syntax and semantics of the AV1 bitstream with the goal to improve interoperability, especially for hardware implementations. The profiles defined in this specification follow the conventions of the [MIAF] specification. Images encoded with [AV1] and not meeting the restrictions of the defined profiles may still be compliant to this AV1 Image File Format if they adhere to the general AVIF requirements.

The AV1 Image File Format supports High Dynamic Range (HDR) and Wide Color Gamut (WCG) images as well as Standard Dynamic Range (SDR). It supports monochrome images as well as multi-channel images with all the bit depths and color spaces specified in [AV1] , and other bit depths with Sample Transform Derived Image Items . The AV1 Image File Format also supports transparency (alpha) and other types of data such as depth maps through auxiliary AV1 bitstreams .

The AV1 Image File Format also supports multi-layer images as specified in [AV1] to be stored both in image items and image sequences. The AV1 Image File Format supports progressive image decoding through layered images.

An AVIF file is designed to be a conformant [HEIF] file for both image items and image sequences. Specifically, this specification follows the recommendations given in "Annex I: Guidelines On Defining New Image Formats and Brands" of [HEIF] .

This specification reuses syntax and semantics used in [AV1-ISOBMFF] .

2. Image Items and properties

2.1. AV1 Image Item

When an item is of type av01 , it is called an AV1 Image Item , and shall obey the following constraints:

• The AV1 Image Item shall be a conformant MIAF image item .
• The AV1 Image Item shall be associated with an AV1ItemConfigurationProperty .

• The content of an AV1 Image Item is called the AV1 Image Item Data and shall obey the following constraints:

  • The AV1 Image Item Data shall be identical to the content of an AV1 Sample marked as ' sync ' , as defined in [AV1-ISOBMFF] .
  • The AV1 Image Item Data shall have exactly one Sequence Header OBU .

    NOTE: File writers may want to set the still_picture and reduced_still_picture_header flags to 1 when possible in the Sequence Header OBU part of the AV1 Image Item Data so that AV1 header overhead is minimized.

2.2. Image Item Properties

2.2.1. AV1 Item Configuration Property

Box Type:                 av1C
Property type:            Descriptive item property
Container:                ItemPropertyContainerBox
Mandatory (per item):     Yes, for an image item of type 'av01', no otherwise
Quantity (per item):      One for an image item of type 'av01', zero otherwise

The syntax and semantics of the AV1ItemConfigurationProperty are identical to those of the AV1CodecConfigurationBox defined in [AV1-ISOBMFF] , with the following constraints:

• Sequence Header OBUs should not be present in the AV1ItemConfigurationProperty .
• If a Sequence Header OBU is present in the AV1ItemConfigurationProperty , it shall match the Sequence Header OBU in the AV1 Image Item Data .
• The values of the fields in the AV1ItemConfigurationProperty shall match those of the Sequence Header OBU in the AV1 Image Item Data .
• The values of the bit depth and the number of channels derived from the AV1ItemConfigurationProperty shall match the PixelInformationProperty ( ' pixi ' ) if present.
• Metadata OBUs , if present, shall match the values given in other item properties , such as the MasteringDisplayColourVolumeBox ( ' mdcv ' ) or ContentLightLevelBox ( ' clli ' ).

2.2.2. Image Spatial Extents Property

The semantics of the ' ispe ' property as defined in [HEIF] apply. More specifically, for [AV1] images, the values of image_width and image_height shall respectively equal the values of UpscaledWidth and FrameHeight as defined in [AV1] but for a specific frame in the item payload. The exact frame depends on the presence and content of the ' lsel ' and OperatingPointSelectorProperty properties as follows:

• In the absence of a ' lsel ' property associated with the item, or if it is present and its layer_id value is set to 0xFFFF:

  • If no OperatingPointSelectorProperty is associated with the item, the ' ispe ' property shall document the dimensions of the last frame decoded when processing the operating point whose index is 0 .

  • If an OperatingPointSelectorProperty is associated with the item, the ' ispe ' property shall document the dimensions of the last frame decoded when processing the corresponding operating point .

  NOTE: The dimensions of possible intermediate output images might not match the ones given in the ' ispe ' property. If renderers display these intermediate images, they are expected to scale the output image to match the ' ispe ' property.

• If a ' lsel ' property is associated with an item and its layer_id is different from 0xFFFF, the ' ispe ' property documents the dimensions of the output frame produced by decoding the corresponding layer.

NOTE: The dimensions indicated in the ' ispe ' property might not match the values max_frame_width_minus1 +1 and max_frame_height_minus1 +1 indicated in the AV1 bitstream.

NOTE: The values of render_width_minus1 and render_height_minus1 possibly present in the AV1 bitstream are not exposed at the AVIF container level.

2.2.3. Clean Aperture Property

The semantics of the clean aperture property ( ' clap ' ) as defined in [HEIF] apply. In addition to the restrictions on transformative item property ordering specified in [MIAF] , the following restriction also applies:

2.2.4. Other Item Properties

In addition to the Image Properties defined in this document, AV1 image items may also be associated with item properties defined in other specifications such as [HEIF] and [MIAF] . Commonly used item properties can be found in § 9.1.1 Minimum set of boxes and § 9.1.2 Requirements on additional image item related boxes .

In general, it is recommended to use item properties instead of Metadata OBUs in the AV1ItemConfigurationProperty .

2.3. AV1 Layered Image Items

2.3.1. Overview

[AV1] supports encoding a frame using multiple spatial layers. A spatial layer may improve the resolution or quality of the image decoded based on one or more of the previous layers. A layer may also provide an image that does not depend on the previous layers. Additionally, not all layers are expected to produce an image meant to be rendered. Some decoded images may be used only as intermediate decodes. Finally, layers are grouped into one or more Operating Points . The Sequence Header OBU defines the list of Operating Points , provides required decoding capabilities, and indicates which layers form each Operating Point .

[AV1] delegates the selection of which Operating Point to process to the application, by means of a function called choose_operating_point() . AVIF defines the OperatingPointSelectorProperty to control this selection. In the absence of an OperatingPointSelectorProperty associated with an AV1 Image Item , the AVIF renderer is free to process any Operating Point present in the AV1 Image Item Data . In particular, when the AV1 Image Item is composed of a unique Operating Point , the OperatingPointSelectorProperty should not be present . If an OperatingPointSelectorProperty is associated with an AV1 Image Item , the op_index field indicates which Operating Point is expected to be processed for this item.

NOTE: When an author wants to offer the ability to render multiple Operating Points from the same AV1 image (e.g. in the case of multi-view images), multiple AV1 Image Items can be created that share the same AV1 Image Item Data but have different OperatingPointSelectorProperties .

[AV1] expects the renderer to display only one frame within the selected Operating Point , which should be the highest spatial layer that is both within the Operating Point and present within the temporal unit, but [AV1] leaves the option for other applications to set their own policy about which frames are output, as defined in the general output process. AVIF sets a different policy, and defines how the ' lsel ' property (mandated by [HEIF] for layered images) is used to control which layer is rendered. According to [HEIF] , the interpretation of the layer_id field in the ' lsel ' property is codec specific. In this specification, the value 0xFFFF is reserved for a special meaning. If a ' lsel ' property is associated with an AV1 Image Item but its layer_id value is set to 0xFFFF, the renderer is free to render either only the output image of the highest spatial layer, or to render all output images of all the intermediate layers and the highest spatial layer, resulting in a form of progressive decoding. If a ' lsel ' property is associated with an AV1 Image Item and the value of layer_id is not 0xFFFF, the renderer is expected to render only the output image for that layer.

NOTE: When such a progressive decoding of the layers within an Operating Point is not desired or when an author wants to expose each layer as a specific item, multiple AV1 Image Items sharing the same AV1 Image Item Data can be created and associated with different ' lsel ' properties, each with a different value of layer_id .

2.3.2. Properties

2.3.2.1. Operating Point Selector Property

2.3.2.1.1. Definition

Box Type:              a1op
Property type:         Descriptive item property
Container:             ItemPropertyContainerBox
Mandatory (per item):  No
Quantity (per item):   Zero or one

2.3.2.1.2. Description

An OperatingPointSelectorProperty may be associated with an AV1 Image Item to provide the index of the operating point to be processed for this item. If associated, it shall be marked as essential.

2.3.2.1.3. Syntax

class OperatingPointSelectorProperty extends ItemProperty('a1op') {
    unsigned int(8) op_index;
}

2.3.2.1.4. Semantics

op_index indicates the index of the operating point to be processed for this item. Its value shall be between 0 and operating_points_cnt_minus_1 inclusive.

2.3.2.2. Layer Selector Property

The ' lsel ' property defined in [HEIF] may be associated with an AV1 Image Item . The layer_id indicates the value of the spatial_id to render. The value shall be between 0 and 3, or the special value 0xFFFF. When a value between 0 and 3 is used, the corresponding spatial layer shall be present in the bitstream and shall produce an output frame . Other layers may be needed to decode the indicated layer. When the special value 0xFFFF is used, progressive decoding is allowed as described in § 2.3.1 Overview .

2.3.2.3. Layered Image Indexing Property

2.3.2.3.1. Definition

Box Type:              a1lx
Property type:         Descriptive item property
Container:             ItemPropertyContainerBox
Mandatory (per item):  No
Quantity (per item):   Zero or one

2.3.2.3.2. Description

The AV1LayeredImageIndexingProperty property may be associated with an AV1 Image Item . It should not be associated with AV1 Image Items consisting of only one layer.

The AV1LayeredImageIndexingProperty documents the size in bytes of each layer (except the last one) in the AV1 Image Item Data , and enables determining the byte ranges required to process one or more layers of an Operating Point . If associated, it shall not be marked as essential.

2.3.2.3.3. Syntax

class AV1LayeredImageIndexingProperty extends ItemProperty('a1lx') {
    unsigned int(7) reserved = 0;
    unsigned int(1) large_size;
    FieldLength = (large_size + 1) * 16;
    unsigned int(FieldLength) layer_size[3];
}

2.3.2.3.4. Semantics

layer_size indicates the number of bytes corresponding to each layer in the item payload, except for the last layer. Values are provided in increasing order of spatial_id . A value of zero means that all the layers except the last one have been documented and following values shall be 0 . The number of non-zero values shall match the number of layers in the image minus one.

NOTE: The size of the last layer can be determined by subtracting the sum of the sizes of all layers indicated in this property from the entire item size.

3. Image Sequences

An AV1 Image Sequence is defined as a set of AV1 Temporal Units stored in an AV1 track as defined in [AV1-ISOBMFF] with the following constraints:

• The track shall be a valid MIAF image sequence .
• The track handler for an AV1 Image Sequence shall be ' pict ' .
• The track shall have only one AV1 Sample description entry.
• If multiple Sequence Header OBUs are present in the track payload, they shall be identical.

4. Other Image Items and Sequences

4.1. Auxiliary Image Items and Sequences

An AV1 Auxiliary Image Item (respectively an AV1 Auxiliary Image Sequence ) is an AV1 Image Item (respectively AV1 Image Sequence ) with the following additional constraints:

• It shall be a compliant MIAF Auxiliary Image Item (respectively MIAF Auxiliary Image Sequence ).
• The mono_chrome field in the Sequence Header OBU shall be set to 1.
• The color_range field in the Sequence Header OBU shall be set to 1.

An AV1 Alpha Image Item (respectively an AV1 Alpha Image Sequence ) is an AV1 Auxiliary Image Item (respectively an AV1 Auxiliary Image Sequence ), and as defined in [MIAF] , with the aux_type field of the AuxiliaryTypeProperty (respectively AuxiliaryTypeInfoBox ) set to urn:mpeg:mpegB:cicp:systems:auxiliary:alpha . An AV1 Alpha Image Item (respectively an AV1 Alpha Image Sequence ) shall be encoded with the same bit depth as the associated master AV1 Image Item (respectively AV1 Image Sequence ).

For AV1 Alpha Image Items and AV1 Alpha Image Sequences , the ColourInformationBox ( ' colr ' ) should be omitted. If present, readers shall ignore it.

An AV1 Depth Image Item (respectively an AV1 Depth Image Sequence ) is an AV1 Auxiliary Image Item (respectively an AV1 Auxiliary Image Sequence ), and as defined in [MIAF] , with the aux_type field of the AuxiliaryTypeProperty (respectively AuxiliaryTypeInfoBox ) set to urn:mpeg:mpegB:cicp:systems:auxiliary:depth .

NOTE: [AV1] supports encoding either 3-component images (whose semantics are given by the matrix_coefficients element), or 1-component images (monochrome). When an image requires a different number of components, multiple auxiliary images may be used, each providing additional component(s), according to the semantics of their aux_type field. In such case, the maximum number of components is restricted by number of possible items in a file, coded on 16 or 32 bits.

4.2. Derived Image Items

4.2.1. Grid Derived Image Item

A grid derived image item ( ' grid ' ) as defined in [HEIF] may be used in an AVIF file .

4.2.2. Tone Map Derived Image Item

A tone map derived image item ( ' tmap ' ) as defined in [HEIF] may be used in an AVIF file . When present, the base image item and the ' tmap ' image item should be grouped together by an ' altr ' (see § 5.1 'altr' group ) entity group as recommended in [HEIF] . When present, the gain map image item should be a hidden image item .

4.2.3. Sample Transform Derived Image Item

With a Sample Transform Derived Image Item , pixels at the same position in multiple input image items can be combined into a single output pixel using basic mathematical operations. This can for example be used to work around codec limitations or for storing alterations to an image as non-destructive residuals. With a Sample Transform Derived Image Item it is possible for AVIF to support 16 or more bits of precision per sample, while still offering backward compatibility through a regular 8 to 12-bit AV1 Image Item containing the most significant bits of each sample.

In these sections, a "sample" refers to the value of a pixel for a given channel.

4.2.3.1. Definition

When a derived image item is of type ' sato ' , it is called a Sample Transform Derived Image Item , and its reconstructed image is formed from a set of input image items, constants and operators .

The input images are specified in the SingleItemTypeReferenceBox or SingleItemTypeReferenceBoxLarge entries of type ' dimg ' for this Sample Transform Derived Image Item within the ItemReferenceBox . The input images are in the same order as specified in these entries. In the SingleItemTypeReferenceBox or SingleItemTypeReferenceBoxLarge of type ' dimg ' , the value of the from_item_ID field identifies the Sample Transform Derived Image Item , and the values of the to_item_ID field identify the input images. There are reference_count input image items as specified by the ItemReferenceBox .

The input image items and the Sample Transform Derived Image Item shall:

• each be associated with a PixelInformationProperty and an ' ispe ' property;

• have the same number of channels as defined by the PixelInformationProperty or AV1ItemConfigurationProperty ;

• have the same chroma subsampling; this may be explicitly defined by one of the above properties for some input image items or the Sample Transform Derived Image Item , and may be implicit for the other items (meaning no property defines chroma subsampling for these items);

• have the same dimensions as defined by the ' ispe ' property;

• have the same color information as defined by the ColourInformationBox properties (or lack thereof).

Each output sample of the Sample Transform Derived Image Item is obtained by evaluating an expression consisting of a series of integer operators and operands . An operand is a constant or a sample from an input image item located at the same channel index and at the same spatial coordinates as the output sample.

No color space conversion, matrix coefficients, or transfer characteristics function shall be applied to the input samples. They are already in the same color space as the output samples.

The output reconstructed image is made up of the output samples, whose values shall each be clamped to fit in the number of bits per sample as defined by the PixelInformationProperty of the reconstructed image item. The full_range_flag field of the ColourInformationBox property of colour_type ' nclx ' also defines a range of values to clamp to, as defined in [CICP] .

NOTE: Appendix A: (informative) Sample Transform Derived Image Item Examples contains examples of Sample Transform Derived Image Item usage.

4.2.3.2. Syntax

An expression is a series of tokens . A token is an operand or an operator . An operand can be a literal constant value or a sample value. A stack is used to keep track of the results of the subexpressions . An operator takes either one or two input operands . Each unary operator pops one value from the stack. Each binary operator pops two values from the stack, the first being the right operand and the second being the left operand . Each token results in a value pushed to the stack. The single remaining value in the stack after evaluating the whole expression is the resulting output sample.

aligned(8) class SampleTransform {
    unsigned int(2) version = 0;
    unsigned int(4) reserved;
    unsigned int(2) bit_depth; // Enum signaling signed 8, 16, 32 or 64-bit.
    // Create an empty stack of signed integer elements of that depth.
    unsigned int(8) token_count;
    for (i=0; i<token_count; i++) {
        unsigned int(8) token;
        if (token == 0) {
            // Push the 'constant' value to the stack.
            signed int(1<<(bit_depth+3)) constant;
        } else if (token <= 32) {
            // Push the sample value from the 'token'th input image item
            // to the stack.
        } else {
            if (token >= 64 && token <= 67) {
                // Unary operator. Pop the operand from the stack.
            } else if (token >= 128 && token <= 137) {
                // Binary operator. Pop the right operand
                // and then the left operand from the stack.
            }
            // Apply operator 'token' and push the result to the stack.
        }
    }
    // Output the single remaining stack element.
}

4.2.3.3. Semantics

version shall be equal to 0. Readers shall ignore a Sample Transform Derived Image Item with an unrecognized version number.

reserved shall be equal to 0. The value of reserved shall be ignored by readers.

bit_depth determines the precision (from 8 to 64 bits, see Table 1 ) of the signed integer temporary variable supporting the intermediate results of the operations. It also determines the precision of the stack elements and the field size of the constant fields. This intermediate precision shall be high enough so that all input sample values fit into that signed bit depth.

The result of any computation underflowing or overflowing the intermediate bit depth is replaced by -2 ^{num_bits -1} and 2 ^{num_bits -1} -1, respectively. Encoder implementations should not create files leading to potential computation underflow or overflow. Decoder implementations shall check for computation underflow or overflow and clamp the results accordingly. Computations with operands of negative values use the two’s-complement representation.

token_count is the expected number of tokens to read. The value of token_count shall be greater than 0.

token determines the type of the operand ( constant or input image item sample) or the operator (how to transform one or two operands into the result). See Table 2 . Readers shall ignore a Sample Transform Derived Image Item with a reserved token value.

constant is a literal signed value extracted from the stream with a precision of intermediate bit depth , pushed to the stack.

4.2.3.4. Constraints

Sample Transform Derived Image Items use the postfix notation to evaluate the result of the whole expression for each reconstructed image item sample.

• The tokens shall be evaluated in the order they are defined in the metadata (the SampleTransform structure defined in § 4.2.3.2 Syntax ) of the Sample Transform Derived Image Item .
• token shall be at most reference_count when evaluating a sample operand (when $1\le \mathrm{token}\le 32$ ).
• There shall be at least one token .

• The stack is empty before evaluating the first token .

• There shall be at least 1 element in the stack before evaluating a unary operator .
• There shall be at least 2 elements in the stack before evaluating a binary operator .
• There shall be exactly one remaining element in the stack after evaluating the last token . This element is the value of the reconstructed image item sample.

Non-compliant expressions shall be rejected by parsers as invalid files.

Note: Because each operator pops one or two elements and then pushes one element to the stack, there is at most one more operand than operators in the expression . There are at least $floor\left(\frac{\mathrm{<a\; data-link-type="dfn"\; href="\#sato-token\_count"\; id="ref-for-sato-token\_count①">\; token\_count\; </a>}}{2}\right)$ operators and at most $ceil\left(\frac{\mathrm{token\_count}}{2}\right)$ operands . token_count is at most 255, meaning the maximum stack size for a valid expression is 128.

5. Entity groups

The GroupsListBox ( ' grpl ' ) defined in [ISOBMFF] may be used to group multiple image items or tracks in a file together. The type of the group describes how the image items or tracks are related. Decoders should ignore groups of unknown type.

5.1. ' altr ' group

The ' altr ' entity group as defined in [ISOBMFF] may be used to mark multiple items or tracks as alternatives to each other. Only one item or track in the ' altr ' group should be played or processed. This grouping is useful for defining a fallback for parsers when new types of items or essential item properties are introduced.

5.2. ' ster ' group

The ' ster ' entity group as defined in [HEIF] may be used to indicate that two image items form a stereo pair suitable for stereoscopic viewing.

6. Brands, Internet media types and file extensions

6.1. Brands overview

As defined by [ISOBMFF] , the presence of a brand in the FileTypeBox can be interpreted as the permission for those AV1 Image File Format readers/parsers and AV1 Image File Format renderers that only implement the features required by the brand, to process the corresponding file and only the parts (e.g. items or sequences) that comply with the brand.

An AV1 Image File Format file may conform to multiple brands. Similarly, an AV1 Image File Format reader/parser or AV1 Image File Format renderer may be capable of processing the features associated with one or more brands.

If any of the brands defined in this document is specified in the major_brand field of the FileTypeBox , the file extension and Internet Media Type should respectively be " .avif " and " image/avif " as defined in § 10 AVIF Media Type Registration .

6.2. AVIF image and image collection brand

Files that indicate this brand in the FileTypeBox shall comply with the following:

• The primary image item shall be an AV1 Image Item or be a derived image that references directly or indirectly one or more items that all are AV1 Image Items .

• AV1 auxiliary image items may be present in the file.

Additionally, the brand avio is defined. If the file indicates the brand avio in the FileTypeBox , then the primary image item or all the items referenced by the primary image item shall be AV1 image items made only of Intra Frames .

6.3. AVIF image sequence brands

Files that indicate this brand in the FileTypeBox shall comply with the following:

• they shall contain one or more AV1 image sequences .

• they may contain AV1 auxiliary image sequences .

Additionally, if a file contains AV1 image sequences and the brand avio is used in the FileTypeBox , the item constraints for this brand shall be met and at least one of the AV1 image sequences shall be made only of AV1 Samples marked as ' sync ' . Conversely, if such a track exists and the constraints of the brand avio on AV1 image items are met, the brand should be used .

NOTE: As defined in [MIAF] , a file that is primarily an image sequence still has at least an image item. Hence, it can also declare brands for signaling the image item.

7. General constraints

The following constraints are common to files compliant with this specification:

• The file shall be compliant with the [MIAF] specification and list ' miaf ' in the FileTypeBox .
• The file shall list ' avif ' or ' avis ' in the FileTypeBox .
• Transformative properties shall not be associated with items in a derivation chain (as defined in [MIAF] ) that serves as an input to a grid derived image item . For example, if a file contains a grid item and its referenced coded image items, cropping, mirroring or rotation transformations are only permitted on the grid item itself.

  NOTE: This constraint further restricts files compared to [MIAF] .

8. Profiles

8.1. Overview

The profiles defined in this section are for enabling interoperability between AV1 Image File Format files and AV1 Image File Format readers/parsers. A profile imposes a set of specific restrictions and is signaled by brands defined in this specification.

It is possible for a file compliant to this AV1 Image File Format to not be able to declare an AVIF profile, if the corresponding AV1 encoding characteristics do not match any of the defined profiles.

NOTE: [AV1] supports 3 bit depths: 8, 10 and 12 bits, and the maximum dimensions of a coded image is 65536x65536, when seq_level_idx is set to 31 (maximum parameters level).

8.2. AVIF Baseline Profile

This section defines the MIAF AV1 Baseline profile of [HEIF] , specifically for [AV1] bitstreams, based on the constraints specified in [MIAF] and identified by the brand MA1B .

If the brand ' MA1B ' is in the FileTypeBox , the common constraints in the section § 6 Brands, Internet media types and file extensions shall apply.

The following shared conditions and requirements from [MIAF] shall apply:

• self-containment (subclause 8.2)

The following shared conditions and requirements from [MIAF] should apply:

• grid-limit (subclause 8.4)
• single-track (subclause 8.5)
• edit-lists (subclause 8.6)
• matched-duration (subclause 8.7)

The following additional constraints apply to all AV1 Image Items and all AV1 Image Sequences :

• The AV1 profile shall be the Main Profile and the level shall be 5.1 or lower.

  NOTE: AV1 tiers are not constrained because timing is optional in image sequences and is not relevant in image items or collections.

  NOTE: Level 5.1 is chosen for the Baseline profile to ensure that no single coded image exceeds 4k resolution, as some decoders may not be able to handle larger images. More precisely, following [AV1] level definitions, coded image items compliant to the AVIF Baseline profile may not have a number of pixels greater than 8912896, a width greater than 8192 or a height greater than 4352. It is still possible to use the Baseline profile to create larger images using a grid derived image item .

8.3. AVIF Advanced Profile

This section defines the MIAF AV1 Advanced profile of [HEIF] , specifically for [AV1] bitstreams, based on the constraints specified in [MIAF] and identified by the brand MA1A .

If the brand ' MA1A ' is in the FileTypeBox , the common constraints in the section § 6 Brands, Internet media types and file extensions shall apply.

The following shared conditions and requirements from [MIAF] shall apply:

• self-containment (subclause 8.2)

The following shared conditions and requirements from [MIAF] should apply:

• grid-limit (subclause 8.4)
• single-track (subclause 8.5)
• edit-lists (subclause 8.6)
• matched-duration (subclause 8.7)

The following additional constraints apply to all AV1 Image Items :

• The AV1 profile shall be the High Profile and the level shall be 6.0 or lower.

  NOTE: Following [AV1] level definitions, coded image items compliant to the AVIF Advanced profile may not have a number of pixels greater than 35651584, a width greater than 16384 or a height greater than 8704. It is still possible to use the Advanced profile to create larger images using a grid derived image item .

The following additional constraints apply only to AV1 Image Sequences :

• The AV1 profile shall be either Main Profile or High Profile.
• The AV1 level for Main Profile shall be 5.1 or lower.
• The AV1 level for High Profile shall be 5.1 or lower.

9. Box requirements

9.1. Image item boxes

9.1.1. Minimum set of boxes

As indicated in § 7 General constraints , an AVIF file is a compliant [MIAF] file. As a consequence, some [ISOBMFF] or [HEIF] boxes are required, as indicated in the following table. The order of the boxes is indicative in the table. The specifications listed in the "Specification" column may require a specific order for a box or for its children and the order shall be respected. For example, per [ISOBMFF] , the FileTypeBox is required to appear first in an AVIF file . The "Version(s)" column in the following table lists the version(s) of the boxes allowed by this brand. With the exception of item properties marked as non-essential, other versions of the boxes shall not be used. "-" means that the box does not have a version.

9.1.2. Requirements on additional image item related boxes

The boxes indicated in the following table may be present in an AVIF file to provide additional signaling for image items. If present, the boxes shall use the version indicated in the table unless the box is an item property marked as non-essential. AVIF readers are expected to understand the boxes and versions listed in this table. The order of the boxes in the table may not be the order of the boxes in the file. Specifications may require a specific order for a box or for its children and the order shall be respected. Additionally, the ' free ' and ' skip ' boxes may be present at any level in the hierarchy and AVIF readers are expected to ignore them. Additional boxes in the ' meta ' hierarchy not listed in the following table may also be present and may be ignored by AVIF readers.

10. AVIF Media Type Registration

The media type "image/avif" is officially registered with IANA and available at: https://www.iana.org/assignments/media-types/image/avif .

11. Changes since v1.1.0 release

• EDITORIAL: Stop using `dfn value` for definitions.

• EDITORIAL: Add assert-ids in the spec for conformance file testing and ComplianceWarden

• Add required list of boxes for AVIF files

• EDITORIAL: Add "per item" to item property definitions

• EDITORIAL: Fix broken link for latest-draft.html

• Relax constraint on transformative properties in derivation chains to only apply to grid items

• Change ispe width to correspond to UpscaledWidth

• Clarify relationship between av1C, metadata OBUs and item properties

• EDITORIAL: Update list of other item properties

• Further clarify relationship between av1C, metadata OBUs and item properties

• Add information on tmap, grpl and altr

• Replace recommendations regarding still picture flags in image items by a note

• Add section 4.2.2 "Sample Transform Derived Image Item"

• Add Appendix A "Sample Transform Derived Image Item Examples"

• Add restriction on usage of clap property

• Adopt MIAF shared constraints

• EDITORIAL: Clean up usage of dfn and linking

• Clarify required versions of non-essential item properties

• EDITORIAL: Add refs, fix wording and format in clauses 1,2

• EDITORIAL: Add sato, alpha, depth, progressive in Scope

• Clean up linking to base specs and duplicated information

• EDITORIAL: Use Markdown syntax coloring for code snippets

• EDITORIAL: Consistently use dfn noexport for derived items

• EDITORIAL: Remove Sample Transform sections from TOC

• EDITORIAL: Add references in the Sample Transform sections

• EDITORIAL: Indent notes as the list items they refer to

• EDITORIAL: Remove inconsistent dots in 9.1.2

• Change structure of optional table of boxes

• Add hidden image item recommendation

• Remove mentions of ftyp compatible_brands

• Remove avio brand recommendation

• EDITORIAL: Fix broken lines in Sample Transform examples

• EDITORIAL: Make assert IDs between profiles unique

• Update the HEIF, ISOBMFF, and MIAF references to the latest versions

• EDITORIAL: Merge asserts for avio brand

• EDITORIAL: Clarify sato item requirements

Appendix A: (informative) Sample Transform Derived Image Item Examples

This informative appendix contains example recipes for extending base AVIF features with Sample Transform Derived Image Items .

Bit depth extension

Sample Transform Derived Image Items allow for more than 12 bits per channel per sample by combining several AV1 image items in multiple ways.

Suffix bit depth extension

The following example describes how to leverage a Sample Transform Derived Image Item on top of a regular 8-bit MIAF image item to extend the decoded bit depth to 16 bits.

Consider the following:

• A MIAF image item being a losslessly coded image item,
  and its PixelInformationProperty with bits_per_channel =8,

• Another image item being a lossily or losslessly coded image item with the same spatial dimensions, the same number of channels, and the same chroma subsampling (or lack thereof) as the first input image item,
  and its PixelInformationProperty with bits_per_channel =8,

• A Sample Transform Derived Image Item with the two items above as input in this order,
  and its PixelInformationProperty with bits_per_channel =16,
  and the following SampleTransform fields:

  • version =0

  • bit_depth =2 (signed 32-bit constant s, stack values and intermediate results)

  • token_count =5

    • token =0, constant =256

    • token =1 (sample from 1 ^st input image item)

    • token =130 (product)

    • token =2 (sample from 2 ^nd input image item)

    • token =128 (sum)

This is equivalent to the following postfix notation (parentheses added for clarity):

This is equivalent to the following infix notation:

Each output sample is equal to the sum of a sample of the first input image item shifted to the left by 8 bits and of a sample of the second input image item. This can be viewed as a bit depth extension of the first input image item by the second input image item. The first input image item contains the 8 most significant bits and the second input image item contains the 8 least significant bits of the 16-bit output reconstructed image item. It is impossible to achieve a bit depth of 16 with a single AV1 image item .

NOTE: If the first input image item is the primary image item and is enclosed in an ' altr ' group (see § 5.1 'altr' group ) with the Sample Transform Derived Image Item , the first input image item is also a backward-compatible 8-bit regular coded image item that can be used by readers that do not support Sample Transform Derived Image Items or do not need extra precision.

NOTE: The second input image item can be marked as hidden to prevent readers from surfacing it to users.

NOTE: The second input image item loses its meaning of the least significant part if any of the most significant bits changes, so the first input image item has to be losslessly encoded. The second input image item supports reasonable loss during encoding.

NOTE: This pattern can be used for reconstructed bit depths beyond 16 by combining more than two input image items or with various input bit depth configurations and operations.

Residual bit depth extension

The following example describes how to leverage a Sample Transform Derived Image Item on top of a regular 12-bit MIAF image item to extend the decoded bit depth to 16 bits.
It differs from the Suffix bit depth extension by its slightly longer series of operations allowing its first input image item to be lossily encoded.

Consider the following:

• A MIAF image item being a lossily coded image item,
  and its PixelInformationProperty with bits_per_channel =12,

• Another image item being a lossily or losslessly coded image item with the same spatial dimensions, the same number of channels, and the same chroma subsampling (or lack thereof) as the first input image item,
  and its PixelInformationProperty with bits_per_channel =8,
  with the following contraints:

• For each sample position in each plane,
  ${\mathrm{sample}}_{\mathrm{original}}$ being the value of the 16-bit original sample at that position in that plane,
  ${\mathrm{sample}}_1$ being the value of the 12-bit sample of the first input image at that position in that plane,
  ${\mathrm{sample}}_2$ being the value of the sample of the second input image at that position in that plane,
  $\approx$ representing similarity within compression loss range,
  • ${\mathrm{sample}}_1\approx \frac{{\mathrm{sample}}_{\mathrm{original}}}{2^4}$
  • ${\mathrm{sample}}_2\approx {\mathrm{sample}}_{\mathrm{original}}-2^4\times {\mathrm{sample}}_1+2^7$
  • $0\le {\mathrm{sample}}_1<2^{12}$
  • $0\le {\mathrm{sample}}_2<2^8$
  • $0\le 2^4\times {\mathrm{sample}}_1+{\mathrm{sample}}_2-2^7<2^{16}$

    NOTE: Files that do not respect this constraint will still decode successfully because Clause § 4.2.3.1 Definition mandates the resulting values to be each clamped to fit in the number of bits per sample as defined by the PixelInformationProperty of the reconstructed image item.

• A Sample Transform Derived Image Item with the two items above as input in this order,
  and its PixelInformationProperty with bits_per_channel =16,
  and the following SampleTransform fields:

  • version =0

  • bit_depth =2 (signed 32-bit constant s, stack values and intermediate results)

  • token_count =7

    • token =0, constant =16

    • token =1 (sample from 1 ^st input image item)

    • token =130 (product)

    • token =2 (sample from 2 ^nd input image item)

    • token =128 (sum)

    • token =0, constant =128

    • token =129 (difference)

This is equivalent to the following postfix notation (parentheses added for clarity):

This is equivalent to the following infix notation:

Each output sample is equal to the sum of a sample of the first input image item shifted to the left by 4 bits and of a sample of the second input image item offset by -128. This can be viewed as a bit depth extension of the first input image item by the second input image item, which contains the residuals to correct the precision loss of the first input image item.

NOTE: If the first input image item is the primary image item and is enclosed in an ' altr ' group (see § 5.1 'altr' group ) with the derived image item, the first input image item is also a backward-compatible 12-bit regular coded image item that can be used by decoding contexts that do not support Sample Transform Derived Image Items or do not need extra precision.

NOTE: The second input image item can be marked as hidden to prevent readers from surfacing it to users.

NOTE: The first input image item supports reasonable loss during encoding because the second input image item "overlaps" by 4 bits to correct the loss. The second input image item supports reasonable loss during encoding.

NOTE: This pattern can be used for reconstructed bit depths beyond 16 by combining more than two input image items or with various input bit depth configurations and operations.