Limited by my level, the test may have errors and incompleteness, which is for reference only.
This can also be used as a reference https://blog.nannan.cool/archives/105/
The test is recorded at https://github.com/ted423/FXXS-Encod...discussions/20 , please be careful not to make irrelevant replies.
There is a problem with the qp test in the test. The theory should be aq-mode 0 to test, but I didn't pay attention when it was measured.
However, I personally feel that it has little impact. Unless there is a more detailed document explaining the role of qp, it is unlikely and necessary to retest.
av1 currently does not have enough noise retention. It can't compare to x264 in high code, and it is not as good as 20Mbps x264GP at 21Mbps (because it is too large, so it is not released, only a comparison picture), see Portrait de la jeune fille en feu AKA Portrait of a Lady on Fire 2019 related tests. In the current test in this regard, x265 has surpassed x264 by a lot.
In the follow-up, the original disk Rubber 2010 with relatively clean picture was selected for testing. Under the picture of 265 10.8Mbps close to x264 13Mbps, AV1 lost a lot of details and noise, but the frame was well preserved.

In the low-code test of Portrait de la jeune fille en feu AKA Portrait of a Lady on Fire 2019, the performance of AV1 is quite good, but the noise reduction of x265 is not very good. AV1 wins). Limited by my level, this test may have great discrepancies.
BTW, when the film-grain is turned on to 8, the moles on the face have completely disappeared.

Therefore, the current conclusion is that AV1 is very suitable for noise reduction below 10Mbps. (mainly in that the parameters can be written directly)
but if it involves the need to preserve noise and details, AV1 is not competent. And due to the needs of AV1's application scenarios, it is likely that AV1 will not carry out this enhancement in the future.