AV1 HDR Floating-Point Color in Mezzanine Workflows

This article examines how the AV1 video codec interfaces with high-dynamic range (HDR) floating-point color spaces within intermediate and mezzanine workflows. Because AV1 natively operates using integer-based bit depths up to 12 bits, utilizing it for high-end production pipelines requires specific transformations. By leveraging standard transfer functions, the AV1 Professional Profile, precise metadata signaling, and Academy Color Encoding System (ACES) color pipeline conversions, production systems effectively map floating-point scene data into highly efficient, high-fidelity mezzanine assets.

The Integer Architecture of AV1

AV1 is fundamentally an integer-based discrete cosine transform (DCT) codec. It defines three primary profiles:

• Main Profile: Supports 8-bit and 10-bit depth with 4:2:0 chroma subsampling.
• High Profile: Supports 8-bit and 10-bit depth with 4:4:4 chroma subsampling.
• Professional Profile: Supports 8-bit, 10-bit, and 12-bit depth with 4:0:0, 4:2:0, 4:2:2, and 4:4:4 chroma subsampling.

Because AV1 lacks a native 16-bit or 32-bit IEEE 754 floating-point sample representation, it cannot store raw floating-point values directly in the bitstream. Instead, mezzanine workflows ingest high-dynamic-range floating-point values—such as those found in OpenEXR, raw camera formats, or ACEScg—by normalizing and quantizing the continuous float data into AV1’s high-precision 10-bit or 12-bit integer container space.

Non-Linear Mapping via Transfer Functions

To preserve the wide dynamic range and extended color volume of floating-point representations without causing visible quantization artifacts (banding), the data must pass through a non-linear transfer function before encoding:

1. Perceptual Quantizer (SMPTE ST 2084 / PQ): In workflows targeting dynamic ranges up to 10,000 nits, linear floating-point values are converted to non-linear integer values using the PQ curve. When combined with AV1’s 10-bit or 12-bit encoding, this non-linear distribution concentrates bits where human visual perception is most sensitive.
2. Hybrid Log-Gamma (Rec. ITU-R BT.2100 / HLG): Used predominantly in broadcast-adjacent mezzanine workflows, HLG applies a logarithmic curve to the upper luminance range while maintaining a standard gamma curve for low-light regions.
3. Logarithmic Intermediate Spaces (e.g., ACEScct, LogC): In multi-generational post-production workflows, scene-linear floating-point data is mapped into a standardized camera log curve. AV1’s 12-bit 4:4:4 format offers enough dynamic overhead to store these log-encoded representations while preserving critical highlight and shadow details for subsequent color grading passes.

Color Metadata Signaling

AV1 maintains color fidelity across mezzanine stages through standardized metadata signaling defined inside Open Bitstream Units (OBUs):

• Coding-Independent Code Points (CICP): The sequence header includes CICP values (per ITU-T H.273) that define color primaries (such as BT.2020), transfer characteristics (such as ST 2084), and matrix coefficients. This ensures that downstream editing and mastering software can accurately invert the non-linear transfer function back into scene-linear floating-point data.
• Mastering Display and Content Light Level Metadata: Metadata OBUs convey mastering display color volume (SMPTE ST 2086) alongside Content Light Level information (MaxCLL and MaxFALL). This guarantees that color values mapped from floating-point remain aligned with target reference monitors.

ACES Integration in AV1 Workflows

In modern production environments using the Academy Color Encoding System, AV1 acts as a lightweight intermediate format. Scene-referred 16-bit half-float data (ACES2065-1 or ACEScg) is passed through an Output Transform or converted to an integer intermediate representation (such as 12-bit integer RGB mapped to ACESproxy or a mastering color space).

When decoded, the integer data is unpacked, un-quantized, and restored to 16-bit or 32-bit floating-point space inside the digital content creation (DCC) software. Provided that the initial quantization utilizes 12-bit 4:4:4 chroma sampling, the perceptual difference between the original floating-point source and the AV1-compressed representation remains minimal, allowing AV1 to serve as a high-compression mezzanine archive or review format.