How AV1 Prevents Color Banding at Low Bitrates

Smooth color gradients, such as clear skies, sunsets, and shadowed backgrounds, are notoriously prone to visible color banding artifacts at low and medium bitrates. The AV1 video codec effectively eliminates this issue without requiring excessive bandwidth through a coordinated set of technologies: native 10-bit color depth, decoder-side Film Grain Synthesis, flexible superblock partitioning, and advanced in-loop filtering. These mechanisms allow encoders to maintain imperceptible transitions across broad, flat areas while keeping the overall data footprint remarkably small.

Native 10-Bit Color Encoding

Banding—also known as contouring—occurs primarily when a video lacks sufficient numerical precision to represent subtle color shifts, forcing continuous shades into abrupt, visible steps. Standard 8-bit video provides only 256 luminance and chrominance values per channel, making banding almost inevitable in gentle gradients.

AV1 integrates 10-bit encoding natively into its Main Profile. By utilizing 10-bit color (1,024 values per channel), AV1 quadruples the available gradations of shade. Crucially, AV1’s compression efficiency ensures that encoding in 10-bit requires very little bitrate overhead compared to 8-bit—and often results in smaller file sizes because the internal prediction filters operate with higher mathematical precision, producing fewer residual errors for the encoder to compress.

Film Grain Synthesis (FGS)

In legacy codecs like H.264 and standard HEVC, the traditional defense against banding is high-frequency noise or dither, which naturally masks color boundaries. However, compressing this noise is computationally expensive and consumes massive amounts of bitrate. When older encoders drop bits, they discard the grain first, exposing the harsh bands beneath.

AV1 solves this with Film Grain Synthesis (FGS). During the encoding process, the encoder strips the high-frequency grain away from the source footage, allowing the smooth underlying gradient to be compressed cleanly at a low bitrate. The encoder then measures the statistical properties of the original grain—such as pattern, size, and intensity—and transmits these characteristics as lightweight metadata. The decoder reconstructs and overlays this synthetic grain onto the video during playback. This dither effectively masks any residual stepping with virtually zero impact on the bitstream size.

Larger Superblocks and Directional Intra Prediction

Older video formats rely on relatively small block sizes (typically 16x16 macroblocks in H.264), requiring an encoder to slice a large, smooth sky into hundreds of tiny blocks. If adjacent blocks quantize slightly differently, visible grid seams and boundary steps appear.

AV1 utilizes superblocks up to 128x128 pixels. A large, uniform gradient can be represented as a single block or a few large partitions, dramatically reducing the number of block transitions. Furthermore, AV1 features 56 directional intra-prediction angles, allowing the encoder to project smooth spatial gradients in virtually any direction across a block with high mathematical accuracy, requiring minimal residual correction data.

Advanced In-Loop Filtering

To ensure that flat regions remain smooth across block edges, AV1 employs a multi-stage filtering pipeline directly inside the encoding loop: