I'm just here to subscribe to the thread... It's a fascinating question and I'm interested in the answer to this as well.

Out of curiosity, since you said you scaled the materials I'm assuming you've shrunk the model down before rendering. I was wondering if you've run a test with scaling the model vs non scaling. Does it make a difference in the transparency?

This is a very strange blur effect that I have not seen so far. Generally the models are not supposed to be scaled according to scrubs. When I render glass or transparency I pay attention to the following things: the glass material must not have dirt and scratches. The scratches can only be rendered visibly with a very high sampling rate (2000+). At a lower sampling rate, the denoiser smears all the details and the material looks dull. If this cannot be avoided then I create a simple new glass material. Otherwise I make sure that a nice area light hits the material to create nice reflections that make the glass stand out.
Hope this helps Cheerz

^That was one of the foxes I was considering. I’m glad you can suggest it—I’ll be trying it asap! To answer your other clarifying questions: I do scale down by 1000% (scale-> 0.001) so that i dont deal with culling in the viewport or render. I am aware I can change all that but I have never had good results from blender with huge models. I then scale down all of the Mecabricks textures by the same amount using a python script that loops through them. However the scratches and dirt is something I have been wondering about. There is no built-in way as I can tell to adjust by part, so do you go through the part materials in the shader nodes and manually disconnect the effects for all clear pieces?

As for denoising, I actually found a super sneaky technique from a youtube video: instead of using any built-in denoisers or even the denoiser node, you render out every light pass, denoise them all individually, and combine with the appropriate mixing. This means the pixel colors given to the denoiser each time exist in much tighter color ranges, mostly ranges of grey, and so the result is much less splotchy. It also means that fine detail textures don’t get destroyed even with lower sample rates. However, for clear parts, my issue is actually before denoising—even the noisy light pass renders are blurry (which is why I suspected caustics might be a key issue).

Edit: It may only be 0.01, I’ve automated it so I dont bother to remember off the top of my head.

Hi Finn,

it is very interesting that you use scaling to avoid culling. I always increase the culling value in the viewport and in the camera. No idea if that makes a difference to the scaling.

As for the scratches and dirt, I use the MB advance plug in. There is a masternode in which u can easily control the values.

The denoising method you use would interest me very much. Maybe you have a link to the yt tutorial. In principle I use the same method. I use e-cycles and the denoiser proceeds in the same or similar way. The individual passes are denoised and joined together. However, the "gloss_indirect" and "gloss_direct" passes need significantly more samples than the difuse passes, at least when many light sources are used.

Hope you find the solution and share it with us 😃 cheerz

Of course, here is the link: https://www.youtube.com/watch?v=lrgXfgKnYag !

In theory, the scale should also effect the light. The inverse square law will remain the same but because lighting power is measured in Watts in Blender, it actually has real-world significance. So it should actually be reasonable to assume that proper setting of units can lead to scientifically perfect lighting for a render of Lego. Of course, that can all be scaled as wished...

I have the advanced plugin too. Are you talking about the control system in the "n" menu or the node built into the shader network?