More about AV1 Film Grain
Visionular has applied the film grain encoding technology that the AV1 standard supports with their Aurora1 AV1 Encoder in a unique way. Traditional encoding techniques can generate uneven grain distribution or even just poor retention of the grain itself. The Aurora1 encoder takes full advantage of the film grain processing capabilities of the AV1 standard resulting in faster encoding times and smaller file sizes without quality loss.
Encode vs. Film Grain: No longer one or the other
You no longer have to choose between a clean encoder and luxurious film grain. In fact, not only is it possible to maintain natural-looking grain while still achieving an efficient encode, but it is also becoming a critical requirement to maintain that authentic film grain, especially when talking about premium content.
Technology

Grain Synthesis
This process removes the grain in the video through a denoising process and estimates the grain parameters. The encoder retains these parameters and then sends them to the decoder along with the compressed video stream. At this point, the grain is re-generated using the estimated parameters and then added back to the reconstructed video.
Grain Preservation
This process re-evaluates and adjusts key algorithms in the video encoder, including Motion Estimation, Intra Prediction, Mode decision, RDO, and post processes. Aurora1’s grain preservation capabilities can make the grains distributed uniformly in one frame and across different frames but with the same original pattern. This is a unique capability and not found in other AV1 encoders.


Preservation + Synthesis
The first two methods have complementary strengths. Grain synthesis is good at synthesizing high-frequency Gaussian noise but struggles with complex or irregular grain. Grain preservation can preserve irregular-difficult-to-preserve fine-grain noise (but it will cost a lot of bits).
Behind The Curtain
Grain Synthesis

This method allows one to preserve the grain while still achieving a lower bit rate and higher quality than directly encoding the grain.
When re-applying the noise, the decoding process uses a 32×32 block which is randomly selected from a 64×64 template. These 32×32 blocks are re-assembled in the same orientation next to each other and can often produce patterns and noticeable artifacts.
Visionular took a creative approach to address this challenge by selecting grain seeds and only using seeds with less obvious patterns for synthesis. The results are a much more natural reconstructed film grain.
Preservation + Synthesis

For complex situations like this, Visionular combines the methods which look something like this:
-
- Decompose the grain.
- Retain and encode the grain that is difficult to generate with the Grain Preservation.
- Use the Grain Synthesis method for the more fine-grain.
In the example below, the grain synthesis method can retain the high-frequency gaussian noise. In contrast, the grain preservation method can retain complex/irregular grains. You end up getting the best of both worlds!

Left: Retains the grain in the encode. Right: Combines both methods of retention and synthesis.
As you can see, combining both methods to address complex grain patterns results in a natural-looking grain that is evenly distributed across the entire frame and fits the original artistic intent of the producer.