[ababak]

Zero roughness backface material shows exactly the same refraction problem.

[eric nixon]

Andrey, I said maybe roughness 0 would help... its hard to say whats gone wrong from that render, you need to setup a test which has some 'realism' to it, otherwise its impossible to tell whats going on. Its also possible that 'fire' cant render this setup.

Ok, here is render with correct refraction nd1.33, and some water drops;

Actually, I don't think assigning a rough glass as the backface material could produce a correct look.

Are you sure?.. the render looks pretty real.

[ababak]

The grapes look very tasty, the drops definitely add extra realism. I'd change the branch material though, shouldn't they be light green?

About the backface material, I am not trying to reproduce your render or exact setup, I tried to find some logic in how it works and how it can be used in the future. Unfortunately for now I don't understand how and when it works in a predictable way

[dk2079]

Its also possible that 'fire' cant render this setup.

well possible, ababak have you compared Fire/external maxwell render?

I recently came across a similar situation where initially it was not so clear what was going wrong:
[solved] Fire Camera does not match external Camera

[ababak]

No, in this case they look identical. My first tests (animated gif above) was from Maxwell Render. I've done quick tests of my Fire screenshots now as well.

[eric nixon]

Small update; I've removed any transparency mapping from the 'rough glass' and changed the thin-sss assymetry to -.3





Tom, I can see in this sunlight test that the ground has patches of shadow where it seems like the caustic (green) light is missing, but that problem would exist even without any backface material. Its the same caustics of caustics problem you can see when applying sharp lighting, to a bunch of balloons, or leaves on a tree.

The test with broad lighting looks believable, no?

Its also remotely possible that this technique only works for spherical shapes?.