AV1@Scale: Film Grain Synthesis, The Awakening

When you watch a high-quality encode that includes the actual noise, there is a startling increase in resolution from seeing a still to seeing the video. The noise is effectively dancing over a signal, and at 24 fps the signal is still perfectly clear behind it.

Whereas if you lossily encode a still that discards the noise and then adds back artificial noise to match the original "aesthetically", the original detail is non-recoverable if this is done frame-by-frame. Watching at 24 fps produces a fundamentally blurrier viewing experience. And it's not subtle -- on old noisy movies the difference in detail can be 2x.

Now, if h.265 or AV1 is actually building its "noise-removed" frames by always taking into account several preceding and following frames while accounting for movement, it could in theory discover the signal of the full detail across time and encode that, and there wouldn't be any loss in detail. But I don't think it does? I'd love to know if I'm mistaken.

But basically, the point is: comparing noise removal and synthesis can't be done using still images. You have to see an actual video comparison side-by-side to determine if detail is being thrown away or preserved. Noise isn't just noise -- noise is detail too.

If you have a few static frames and average them, you improve SNR by retaining the unchanged signal and having the purely random noise cancel itself out. Retaining noise itself is not useful.

I suspect the effect you might be seeing is either just an aesthetic preference for the original grain behavior, or that you are comparing low bandwidth content with heavy compression artifacts like smoothing/low pass filtering (not storing fine detail saves significant bandwidth) to high bandwidth versions that maintain full detail, entirely unrelated to the grain overlaid on top.

That's exactly the point of GP though. Even though each individual frame might be almost indistinguishable from random noise you can still extract patterns over time. This is also the case if you don't average the frames in software but let the viewer's brain do it. If you just remove all "noise" from each frame and then add random noise back those patterns will be lost.

In practice you won't have static frames but also movement so recovering the signal from the noise becomes a lot more complicated.