In T69770#813352, @Lukas Stockner (lukasstockner97) wrote:

This is not a bug.

The Albedo channel was designed for the original Cycles denoiser, which is completely invariant to constant shifts in feature passes. As you all noted, the Albedo pass works just fine with the original Denoiser.

Apparently the same is not true for OpenImageDenoise. In this case, we should adapt the pass to work better with it, but this is a feature request, not a bug - the channel does exactly what it was supposed to do, provide an albedo input that works fine with the original denoiser.

In T69770#813349, @Wo!262 (wo262) wrote:

I can confirm as well, I went through the original Intel documentation and was surprised at how good results they have on their page, and the claim that Denoising Albedo is the best input for the denoiser to have. The current denoiser albedo pass is simply adding the Glossy Color pass to all the others Color passes, but the Glossy Color pass doesn't contain percentage of reflectivity for things with frenel. The intel documentation doesn't specify a correct solution though, it's up to you what passes you choose and what combination, for example, if you want to add reflections to the mix or not, and how to do it, taking just the color or the actual first bounce. My current setup is to add Diffuse Color, SSS color and Translucent color, leaving Glossy color out.

I have the same problem. See that in the render with cycles there are more details, and with the intel denoise we lost the light