AVIF irot Box vs EXIF Orientation Metadata

When an AVIF file contains both an irot (image rotation) box and EXIF orientation metadata, the AVIF specification mandates that the container-level irot property takes precedence, requiring conforming decoders to ignore the EXIF orientation tag. This mechanism ensures consistent image presentation across platforms while preventing "double-rotation" errors, where an image is rotated twice by different layers of the decoding pipeline.

The irot box is an ISOBMFF (ISO Base Media File Format) transformative property that instructs the presentation engine to rotate the reconstructed image grid counter-clockwise by 0, 90, 180, or 270 degrees. Because AVIF is governed by the Multi-Image Application Format (MIAF / ISO/IEC 23000-22) specifications, geometric transformations applied at the container level represent the canonical visual orientation of the image.

Conversely, EXIF orientation metadata exists as a payload inside a separate metadata item (Exif item) within the container. Tag 0x0112 in EXIF data historically conveys camera sensor orientation. Because legacy image formats relied entirely on this tag, many image viewers and decoders parse EXIF data before rendering.

To resolve the conflict between the two directives, the MIAF specification dictates strict precedence and encoding rules:

1. Decoder Precedence: Conforming AVIF decoders must apply the transformation defined by the irot box (and imir for mirroring, if present) to render the visual presentation. The decoder must ignore the EXIF orientation tag during display to prevent applying the transformation twice.
2. Encoder Requirements: Specifications state that file authors and encoders should normalize the EXIF orientation tag to 1 (indicating Top-Left, or no rotation) whenever an irot or imir property is declared. Alternatively, encoders may strip the EXIF orientation field entirely to maintain single-source-of-truth metadata.

Despite the standard, implementation discrepancies can occur in non-compliant software. If an outdated or naive viewer extracts the raw AV1 bitstream and the EXIF data while ignoring ISOBMFF transformative properties, the image may appear unrotated or misoriented. Conversely, if a viewer processes both the irot box and the EXIF orientation tag without checking specification constraints, the visual output will suffer from an unintended compound rotation. Standard-compliant modern browsers and media frameworks adhere to the specification by rendering strictly according to the irot box and discarding the EXIF orientation value.