Recommended approach to 5mil + poly walkthrough.

[re:FORM]

Getting back on topic.... looking at that animation I posted there are two things to note:

The GI seems to start off a bit rough (watch the GI-shadow under the tiles on the wall) for the first few frames.

There is some kind of corruption/over-brightness on the top leaf of the tree on the left. Any ideas what's causing this? It appears to be reducing as more of the GI is calculated.

[Clifton Santiago]

I noticed the overbright leaves. Maybe try Reinhard color mapping? Maybe 0.5 burn.

[percydaman]

are you using 2 sided material for the leaves? I think for this scene you might be better off nixing the LC, and going with QMC for second bounce. Maybe try a fairly blurred (smooth) irrad map and edge enhancement, if you have the time.

[re:FORM]

'nixing' ?

[percydaman]

nixing=removing

[re:FORM]

nix 1 Pronunciation (nks)
n. Mythology
A water sprite of German mythology, usually in human form or half-human and half-fish.
[German, from Middle High German nickes, from Old High German nihhus.]
nix 2 Pronunciation (nks) Slang
n.
Nothing.
adv.
Not so; no.
tr.v. nixed, nix·ing, nix·es
To forbid, refuse, or veto: Congress nixed the tax hike.

[percydaman]

[re:FORM]

I tried precalcing the LC, but with 3000 samples its crashing. So precalcing is out of the question I recon.

3000 samples
0.01m radius
scale world

Given the memory requirements of the scene, I think the only way is to render the LC frame to frame, and to use multiframe incremenetal for the IR..... unless I am mistaken?

Clifton, did you precalc your IR and LC passes? ie as per vlado's tutorial rendering LC first, then IR using the LC ? or are you doing it all in one pass?

[re:FORM]

Ok, I tried the other form of precalcing IR/LC

IR at medium
LC 1000 0.02 screen
render every 5th frame

Looking at the early results this looks to be a winner. Not only does it look smooth as... it seems to be rendering succesfully on the render farm (even on the lowly 1Gb machine)

Happy days.

[Clifton Santiago]

Sounds good. Screen scale works well for exteriors, and I seem to get smoother results than world as well.

[DaForce]

World mode is mainly used when LC is set to flythru. Which is what you should be doing.

[ene.xis]

ur sample size was way too small its my guess...

I found that in order to get an eficient LC solution when using world scale, i had to increase sample size significantly from what i d imagine inicially...

i d be wiching to find an eficient rule thumb either for the sample size and number...

[re:FORM]

Would it be correct to say that its pointless having a sample size larger than the leaves in scene, as the leaves would end up with 1 or less samples. In that case it would be down to the IR pass to pick up any GI on the leaves? Or am I making incorrect assumptions?

[ene.xis]

i d thought so, but i m not sure really...

nevertheless i guess u shoudn t calculate ur GI on a every frame base...ur wasting rendering time as i see it...

[re:FORM]

hehe, I know... but this scene crashes with LC calculating for the entire walkthrough.
Vlado, is my assumption above correct?