When is regenerating the GI-Caches Necessary?

[PabloMack]

I have been using GI caching a lot to mitigate flicker. Sometimes, though, I have to work on my models because there are gremlins in the data structures and it takes some massaging to work them out. So I end up using the same caches for many different versions of the scene. This makes me wonder what might be good reasons for re-computing the GI-caches. The rule I have been going on is that if the geometry changes, then the caches need to be recomputed. But how about if colors are changed? Do "textures" affect how what goes into the caches or does that only happen when the actual render is done? Asked another way: Do the GI-caches only contain geometric information or do they actually sample lighting including color? Does anyone know?

[Matt]

If the changed textures will cause the light bouncing onto other objects to change colour, then the GI cache should ideally be regenerated. But if you are only concerned about the GI being rendered onto to the changed textures, then the old GI cache would still be valid because it only records the light that illuminates the surface and is not multiplied by the colour of the surface.

Geometry changes should require GI caches to be regenerated in all cases, but often you can get away without doing that if the changes are small.

Matt

[PabloMack]

Sounds like my understanding wasn't too far off. I have some related questions.

1. Are both GI sampling density and rendering quality directly controlled by the "Detail" parameter in the Quality tab of the render node?
2. If I generate GI cache at a different Detail setting than I render at, will it cause a problem with the software or will rendering use all of the GI cache resolution available in the cache files regardless of the Detail setting when rendering?
3. As generating GI cache is much faster than rendering, will it help any to generate GI cache at a higher Detail setting than I render at or is this pointless?

I'm just trying to understand how the Detail parameter used during GI cache generation vs. rendering because I can do these two operations with different Detail settings.

Thanks a bunch.

[Oshyan]

1: Yes in Terragen 2, but no in Terragen 3. In Terragen 3 we decoupled GI prepass detail from main render detail. That's why it used to be called "relative detail" in TG2 and is now just called "GI cache detail".
2: It will not cause problems. Rendering will just use the cache as it is. You can generate a cache at higher or lower detail, or at higher or lower resolution. You can potentially use this for some functional purposes, although in practice simply adjusting the GI Cache Detail setting is generally easier and able to achieve similar goals.
3: The GI generation will be based on the cache you generate, regardless of your main render Detail setting. But it will be applied to the more limited detail of the render if you reduce main render Detail, so the results won't necessarily be better. Of course generating a higher quality GI cache takes longer anyway. But since the settings are now decoupled, you are free to use lower main detail with higher GI cache detail if you want.

As I understand the GI cache it's it's kind of the opposite of what you stated above: the GI cache *is* sampling lighting/color, and while it does store some geometry info about the position and normal of each sample, it is not storing geometry per se. Matt may chime in again with better info.

- Oshyan

[Matt]

Just to clarify something, in Terragen 3 the Detail parameter has absolutely no effect on the GI prepass or the data contained in the GI cache. Even the micropolygon size in the prepass is controlled by the GI cache detail parameter, and only that parameter.

Matt