Guide

AV1 Film Grain Synthesis Explained

By Dirk Hildebrandt, CTO of Wavelet Beam

Film grain is part of the look of a film, and it is one of the most expensive things to encode. AV1 film grain synthesis solves this: the grain is removed before encoding, described in a few bytes of metadata and added back on playback. This guide explains how it works, why it saves so much bitrate and what goes wrong in practice.

Why grain is expensive

Grain is random. It changes in every frame and has no temporal redundancy, so the encoder cannot predict it from previous frames. Block-based codecs such as H.264, HEVC and VVC quantize the high frequencies where grain lives. At low bitrates the grain turns into smooth, plastic areas or blocky noise. At high bitrates it survives, but eats a large part of the budget. Either the look or the bandwidth suffers.

How film grain synthesis works

1. Denoise: the grain is removed from the source before encoding.
2. Estimate: the grain of the original is analyzed and described as parameters, the film grain table: strength per brightness level, grain size and correlation.
3. Encode: the clean picture is encoded. The encoder spends its bits on real image detail.
4. Signal: the grain table travels with the stream as lightweight metadata.
5. Synthesize: the decoder generates new grain from the table and adds it to the decoded picture on playback.

AV1 is the first codec in which film grain synthesis is a mandatory part of the decoder, so every compliant AV1 player can reproduce the grain. In VVC it is optional.

How much bitrate it saves

Each new codec generation is marketed with 50% savings over the previous one. Real-world gains between generations are closer to 20 to 25%. Reductions of 50% or more are practically only reachable with film grain synthesis. Netflix reports around 30% savings with AV1 film grain synthesis, and we confirm this. We encoded the Netflix test scene "Meridian" in UHD with AV1 and film grain synthesis at under 6 Mbit/s: 12 minutes at just 536 MB.

Where it goes wrong: denoising and banding

Film grain synthesis is only as good as the denoising in step 1. Standard denoisers lose fidelity: they smooth fine detail together with the grain. Encoded at low bitrates, the smoothed gradients break into bands. A common workaround is to add more artificial grain to hide the banding. We see it differently: if banding occurs, encoding standards were not maintained, and blanket re-graining does not improve the actual picture.

The problem is well known. At Demuxed 2025, Netflix explained that FGS had to be disabled on critical sequences because standard pre-filtering degraded the image. This causes a visible break for the viewer: the continuous grain experience changes across the timeline from one scene to the next. Generating accurate film grain tables is complex and rarely works out of the box.

How to measure FGS correctly

Synthesized grain is random, so it never matches the original grain pixel by pixel. Measured against the grainy original, every metric punishes the new grain as an error. To judge the true compression efficiency, compare the denoised master against the decoded AV1 stream without re-graining. This way you measure compression quality, not noise.

Look at stability, not only averages. In an AV1 4K encode at 6.696 Mbit/s we reached a VMAF mean of 92.80 with a standard deviation of 1.63 and a PSNR-Y mean of 36.15 dB with a standard deviation of 0.43 dB. A low standard deviation means a calm picture without quality pumping. See our guide on VMAF pooling and QBVE.

Our approach

We combine three building blocks into one pipeline:

• IRIS-Denoising extracts the true signal instead of smoothing it, so there is nothing to hide with extra grain.
• Our own film grain tables, developed from scratch, avoid repeating grain patterns and reproduce the grain of the original as closely as possible.
• Single-pass per-shot encoding gives every shot the bits it needs, exported for HLS and DASH.

The result is UHD cinema aesthetics at bandwidths usually reserved for 1080p. For clean digital footage without grain, IRIS pre-processing alone already improves compressibility for any codec, see why denoise before encoding. More on the AV1 film grain synthesis page.