World Machine Test

[Gannaingh]

I'm trying to use World Machine to create realistic terrain bases and to be used in terragen. This is just the first test that I'm doing to see if I can get a reasonable end result. Next I'm probably going to go with a larger scale landscape so that more varied terrain features are visible.

[Oshyan]

Looks promising, but everything is oddly evenly lit? Like all the objects are highly translucent, or even slightly luminous. Also, curious to see more of the terrain.

- Oshyan

[Gannaingh]

I agree, the trees in particular look weird. I think it's a product of the translucency of the leaves and the fill light setup that i used to try and get some detail in the shadowed areas of the steep slopes. The next one will be a much bigger scene, about 11 km end to end.

[TheBadger]

^^Looking forward to it!

[Oshyan]

TG3 might help you with the detail in shadows thing... Personally I find fill lights to be a thing of the past, GI with new GISD produces superior results every time, and with similar render times and more accuracy. I have converted a number of fill light projects to GI over the past few months and always pleased with the end result. It's kind of amazing what you take for granted as being "correct" (or at least passably realistic) with fill lights that then looks totally different when using proper GI (and better with GI at that). Whatever you do, just don't use shadows on your fills.

- Oshyan

[Tangled-Universe]

I have found out that basically all grass models have incorrectly configured translucency settings.
They are often set to values around 0.9, which is crazy high.

However, I know Walli mostly did this in response to feedback/critique that many of his models look too dark, which as a matter of fact rather is a TG problem than the problem of the model. Without going into too much detail it generally is because of that TG's raytracer "resolution" or accuracy isn't high enough to catch all the individual aspects of the model when it's at greater distance from the camera, which very often results in very dark equally shaded results, because it looks like the renderer chooses to use an average result for the entire silhouette of the model.
The best examples are Silva3D Fir's and Pines which have dense needles. Renders great in every renderer except for TG.
Or another example is NWDA dry bushes. Render these from high'ish altitude so that they are only a few pixels large => all near black or at least all flat shaded without any detail.

A workaround, kind of, was to fiddle with translucency to avoid getting too dark results, because of the "under-estimation".
Consequently, if the screenspace does offer enough visible geometry, like here, then the renderer just renders it exactly as configured and that's what we're seeing here.

So the new GISD in TG3 may offer some improvement on some of these issues.
I'm tied to the NDA of course, so I can't be exact and explicit on how and why at this point.
But Oshyan is right that it might come in useful.

In this specific situation I won't blame the renderer, but rather the model's translucency settings.
I set mine at around 0.2 for grasses and 0.4 max for trees and such. For evergreens I use lower.

For years I tried to "catch" all surface and sky details in one shot, but I'm abandoning that approach now.
It's not photorealistic and too HDR'ish.
I think that's what Oshyan is referring to with flat lighting as well (besides that it also seems back-lit).

Fill lights are seldomly necessary indeed. If you want to use some to get more detailed shadows or trying to accentuate some colours then exploring the enviro light node would be the best suggestion I can give you.

Basically all my latest renders have tweaked enviro light settings + 1 sunlight where I try to choose to expose on either surface or sky.

Sorry for this bit incoherent story, I hope you can distill some useful info from it

Cheers,
Martin

[TheBadger]

^^ Great info guys!!! Thank you.

Also

Personally I find fill lights to be a thing of the past

I just wet my self. Im very happy about this... About the fill lights I mean. Wetting my self makes me feel a little sad.

[Tangled-Universe]

Comparing fill lights only with GI seems ok to me.
Comparing fill lights with GISD seems kind of ok to me as long as GI is enabled with the fill lights.*
Comparing fill lights only with GISD would not make sense though.*

* this is a difficult comparison because of how GISD works in TG2 and how TG3 manages to do this even quicker than TG2.**
A fill light does not do the same as enabling GISD. You can try to simulate GISD with fill lights, but it will only add more direct light which potentially allows the GI to work with more light, but it will not force the renderer to shoot additional rays from each point to gather more detailed lighting of its nearby surroundings, as GISD generally does. They are different and thus not suitable for comparison.
Therefore comparing a fill lights only setup vs GISD (GI on steroids)(TG2 or TG3, doesn't matter) isn't correct.
I'm pretty sure though that Oshyan referred to comparing fill lights along with GI vs GISD.
It's an understandable comparison, but not entirely correct.

**The only correct and honest comparison ((in respect to the features) is TG2 GISD vs TG3 GISD.
This provides a side by side comparison of the two, which is the most simple and effective comparison.
Hence, it is being referred to as "the new GISD".

When doing this it still means that TG3 GISD is better, with more control and much much faster than TG2's GISD.

However, I think it's wise to keep comparisons as simple as possible to make sure the conclusions are correct, to provide the right nuance, but mostly also to make sure everybody understand what's going on

[Oshyan]

I don't think it's always best to view it so technically, though I certainly understand the motivation (I'm generally a technically-oriented person myself). Results are ultimately what matters in my opinion, and I believe people often use fill lights to achieve some of the results that the new GISD method obtains. For example, if you look at Jeff Boser's road scene with lots of vegetation, you can see that there is in fact both more light (highlights) *and* shadows in the GISD version:
http://terragen3.com/focus-on-features-global-illumination-enhancements/

People add fill lights, often with shadows *enabled*, to achieve similar effects (I've seen it numerous times in scenes that are sent to me for troubleshooting and other purposes). To lighten up shadowed areas and add some detail, or at least allow you to see detail that is in shadow, even if it's not really "adding" much.

So it's that kind of thing I'm referring to most particularly. Sure, you can't extrapolate these results to all GI vs. fill light situations and comparisons, fill lights are still useful, etc, etc. But in general I find fill light setups much less necessary now that A: GI is pretty fast, and B: GISD adds significant detail in shadowed areas without a big render time hit. That's a clear correlation in my experience, so I think it's worth mentioning. It may not be the most technically accurate comparison, but it is valuable from a creative perspective in my opinion.

- Oshyan

[Tangled-Universe]

Yes you're right Oshyan.
Result matters most and indeed the approach shouldn't be too technical, although I'm happy that you seem to agree with me about it from that technical perspective. In that perspective any comparison is ok, but performance wise conclusions should be avoided then.

[Matt]

Without going into too much detail it generally is because of that TG's raytracer "resolution" or accuracy isn't high enough to catch all the individual aspects of the model when it's at greater distance from the camera, which very often results in very dark equally shaded results, because it looks like the renderer chooses to use an average result for the entire silhouette of the model.
The best examples are Silva3D Fir's and Pines which have dense needles. Renders great in every renderer except for TG.
Or another example is NWDA dry bushes. Render these from high'ish altitude so that they are only a few pixels large => all near black or at least all flat shaded without any detail.

Do you mean the direct lighting or the GI? If you mean direct lighting, I don't really understand what you mean by a limitation of the resolution or accuracy. With "max samples" on the AA the direct lighting solution should be an unbiased estimate. GI is a different story though. The GI cache averages the GI quite a bit, so in many places it will be underestimated, but this is what GISD in TG3 is designed to help with. You should allow the GI cache to be generated while GISD is enabled to produce a brighter, less biased result.

Matt

[Tangled-Universe]

Thanks Matt,

I'm afraid I have derailed this topic now, I'm sorry for that.

Matt, I'm happy to show you examples of what I mean as it's difficult to explain.

I can't explain it differently or better at the moment than that the renderer renders all the leafs individually with its unique shading due to lighting etc. when the model is relatively close or "large in image space", but when it is further away or "small in image space" it then seemingly treats all the leafs as one and gives each leaf the exact same dark shading. As if it decides something like "nah, too far way, one shade for all the leafs will represent it nicely enough at that distance".

I had a crack at fixing it with Jon (Hetzen) and he concluded it's GI, but upping GI didn't work, nor was GISD satisfactory.
The workaround was doubling to quadrupling the resolution of the render and thus effectively increasing the absolute size of your to be rendered objects.
This is why I used the "resolution" term, see?

Perhaps I should revisit that scene now?
Anyway, if you're still interested to see what I mean then I can hand it over on the alpha forums and leave this topic alone

Cheers,
Martin

[Matt]

Hi Martin,

Yes, I'd be interested to see some results and we can try to find out what's going on. It is to be expected that GI is not accurate at small scales due to the use of the GI cache, but I would also expect that increasing GI relative detail would theoretically have a similar effect to increasing image size.

It's fine to post them here on this public forum too.

Matt

[Gannaingh]

There's some good information in here, thank you all for that. I'll definitely use your suggestions in adjusting my work flow. I'm very excited for Terragen 3, it looks amazing and I can't wait to get my hands on it 

[efflux]

I have found out that basically all grass models have incorrectly configured translucency settings.
They are often set to values around 0.9, which is crazy high.

However, I know Walli mostly did this in response to feedback/critique that many of his models look too dark, which as a matter of fact rather is a TG problem than the problem of the model. Without going into too much detail it generally is because of that TG's raytracer "resolution" or accuracy isn't high enough to catch all the individual aspects of the model when it's at greater distance from the camera, which very often results in very dark equally shaded results, because it looks like the renderer chooses to use an average result for the entire silhouette of the model.
The best examples are Silva3D Fir's and Pines which have dense needles. Renders great in every renderer except for TG.
Or another example is NWDA dry bushes. Render these from high'ish altitude so that they are only a few pixels large => all near black or at least all flat shaded without any detail.

A workaround, kind of, was to fiddle with translucency to avoid getting too dark results, because of the "under-estimation".
Consequently, if the screenspace does offer enough visible geometry, like here, then the renderer just renders it exactly as configured and that's what we're seeing here.

So the new GISD in TG3 may offer some improvement on some of these issues.
I'm tied to the NDA of course, so I can't be exact and explicit on how and why at this point.
But Oshyan is right that it might come in useful.

In this specific situation I won't blame the renderer, but rather the model's translucency settings.
I set mine at around 0.2 for grasses and 0.4 max for trees and such. For evergreens I use lower.

For years I tried to "catch" all surface and sky details in one shot, but I'm abandoning that approach now.
It's not photorealistic and too HDR'ish.
I think that's what Oshyan is referring to with flat lighting as well (besides that it also seems back-lit).

Fill lights are seldomly necessary indeed. If you want to use some to get more detailed shadows or trying to accentuate some colours then exploring the enviro light node would be the best suggestion I can give you.

Basically all my latest renders have tweaked enviro light settings + 1 sunlight where I try to choose to expose on either surface or sky.

Sorry for this bit incoherent story, I hope you can distill some useful info from it

Cheers,
Martin

This is why I never add any trees. First time I tried it they looked like crap compared to how they look in most other apps. Maybe things are better now and maybe Terragen 3 will be better.