Flaw in lightgroup implementation

Hi devs, I’ ve tried the long awaited lightgroup feature and here are my 2 cent.

The result works amazing, as expected. I was one of those asking for this many years ago (since it first appeared in Maxw*ll).
But in the meantime denoising happened, and new workflows happened too.
Lately the best render results come from multi-stage denoising passes and joining them back into final image.
This is not possible using current lightgroups, as they are delivered as single “beauty” images, each to be denoised as a whole.
As it is now the feature is cool but (at least in my work) relegated to lighting tests, instead of final shots compositing.
What I expected was something like a Viewlayer control level, as in this quick mockup:

Is this too late / too difficult to achieve? (or maybe not needed / desirable )

The problem with this implementation is that to access every light group you would need to have image node per light groups, while with what we have right now every light group it’s just a pass, so you just need one image node to compose the whole render.

At first sight I like the idea, at second thought I’m not so sure that it’s better than the actual implementation or it will make a bigger mess in the compositor node tree.

About if its too difficult, I’m not sure, if it’s too late, I don’t think so, maybe for 3.3 cannot be done, but if it’s something accepted it may go for a later release.

In any case I would define this as a low priority target, there are other features, like multiple environments, that I think are way more important :slight_smile:

What do you think about this @lukasstockner97 @brecht @boberfly ?

1 Like

Well the whole point I’m making is that currently the best denoising strategy (waaay best) is to work on separating the passes, denoising, and joining back. The day OIDN denoises as good as this, I’ll wipe this post.
As for the mess in the node tree I wouldn’t care, it is often messy, by its nature I’d say. The denoising step can be done with nodegroups anyway, so it would be the matter of adding as many renderlayers as needed.

1 Like

I usually denoise in the compositor, but in general we need to be able to enable render-time denoise for some passes, specially for something like the light groups, I agree with you on this for sure.

But I think I got what you mean, correct me if I’m wrong, but you mean getting all the render passes for every light groups, like diffuse direct/indirect/color and such, right?

I agree that it would be a very welcome addition for many many reasons, I think @lukasstockner97 had that in his list of improvements for lightgroups, but there was some complications, I’m not entirely sure.

Jajaja, you’re right.
I probably had explanation problems before :slight_smile: