[lifeofdave]

Hi all

In an Arch Viz scene I'm working on I'm trying to get some light to bounce off the surrounding glass buildings onto the pavement and new building.

This is the kind of effect I'm after:


The approach so far has been to include a large and very reflective plain with a smooth and bumpy normal map to bounce the sunlight onto the model whilst adding the colours of the surrounding building.

Here is the scene setup, current results and material properties:



I'm happy with the effect the setup is producing, but need to reduce the noise.

Anyone have any ideas on how to produce this effect with a relatively low noise level without going to a v.high sl?

thanks in advance for any suggestions!

Dave

[dariolanza]

Hello lifeofdave,

I guess that this noise and the slow performance is being produced by the Bump-Normal map in the reflective object.

In fact, the effect you trying to get is caustic light, reflected on the glass window panes of the surrounding buildings. And glass window panes are not bumpy or rough, but smooth and planar.

So, I would try removing any bump effect in the reflective building (the one behind the camera) and see if the caustics are getting cleaner and faster.

If this test goes fine, try using a bit of Displacement (smooth and vague) instead of bump-normal, to get an slightly deformed and not planar building (to get slightly distorted caustics).

Let me know.

Dario Lanza

[Bubbaloo]

Good advice Dario. I would also suggest using a metal material as the reflective plane. Subdivide the plane and give it a small amount of noise to make it not perfectly flat.

[lifeofdave]

Dario, thanks for your reply.

I've done some further tests, starting with a more basic scene and simple chrome reflector with no bump etc, here are the results:



The first 3 tests are in a simple scene, all emitters, glass etc stripped out. Only default and very basic materials left.
All test are rendered to around sl 18.
1) Simple pink chrome plain
2) Chrome plain with ref 0 and ref 90 reflection photo
3) Chrome plain with ref 0 and ref 90 reflection photo with smooth undulating normal map to create a caustic like effect.

4) 4th test uses the same plain material settings and environment as number 3, however I have reintroduced all the other textures and emitters.

As soon as the complexity of the scene is increased (test no.4) in terms of other textures and emitters the strength of the reflected light at sl 18 is much weaker and noisier, it looks as if the reflected light needs to go to a much higher sampe level than in the basic tests . Is this expected behaviour? I'm still learning a lot about what a sample level really represents and emitter priority etc...

The addition of the normal map in test 3 didn't increase render time or noise significantly, so I haven't experimented with displacement yet as it's something I'm not familiar with, but I will definitely have a play sometime soon.

I'm going to leave this scene rendering over the weekend so should be able to see how things clear up at a much higher sl and resolution.

[lifeofdave]

Cheers Bubbaloo, I'll give that a shot as well when I get a moment.

[Mihai]

Probably to do with how emitter strengths are optimized. You are using Physical Sky sun together with normal emitters? If you have very strong and normal emitters in the same scene, there has to be a balance between the light calculated from the strongest vs weaker emitters. This makes it so the caustics from the sun will be calculated slower when you add the regular emitters in. I *think* this is what's happening. Ideally the emitter strengths should be balanced and you can always lower the ones you need to be weaker or make others stronger using ML.

[lifeofdave]

Mihai, that helps a lot. I'm trying to avoid mulitlight for the final render as it's 5k pixels wide and there are so many emitters that the ml mxi is huge and rendering uses far more ram than our nodes have. In the end I did a limited region render for the reflected light effect with all other emitters turned off and that's worked out well. cheers for all your advice.

[numerobis]

Probably to do with how emitter strengths are optimized. You are using Physical Sky sun together with normal emitters? If you have very strong and normal emitters in the same scene, there has to be a balance between the light calculated from the strongest vs weaker emitters. This makes it so the caustics from the sun will be calculated slower when you add the regular emitters in. I *think* this is what's happening. Ideally the emitter strengths should be balanced and you can always lower the ones you need to be weaker or make others stronger using ML.

if this leveling is really working, why cant this be managed internally by the engine?!?

[feynman]

If it's still images, couldn't one just use a projector, meaning, a textured emitter?

[lifeofdave]

If it's still images, couldn't one just use a projector, meaning, a textured emitter?

Are you suggesting using a textured emitter on the reflection plain? This does add some colour variance across the ground and building but it's very gradual and subtle. Not like the strong, caustic reflected light spots created by glass buildings. One approach I haven't tried is using textured emitters that are really close to the building and ground surface, say 100mm. This might work but creating a texture that emits a caustic-like light that resembles the surrounding context is another mission in itself.

[feynman]

I was thinking of taking some HDR photo and mapping that onto a plane which is invisible to camera, etc. If you can't see the buildings that produce the reflections, maybe that's one quick way to fake what you're after? Just an idea, maybe not practical.

[eric nixon]

Is your 'simple chrome reflector' roughness 0 ? - It should be, that will speed up caustics greatly.
A smooth 16bit undulating normal map - mimicing the glazed facade surface which the sunlight is bouncing off.
Rather than use maxwell sun I would use an emitter 500m away + multilight...

Dario said something along those lines about a smooth undulating surface.

[Fernando Tella]

Maybe you can make a pass just for this with no other emitters or lights and add it in photoshop to the main pass in linear dodge mode.

[Mihai]

Mihai, that helps a lot. I'm trying to avoid mulitlight for the final render as it's 5k pixels wide and there are so many emitters that the ml mxi is huge and rendering uses far more ram than our nodes have. In the end I did a limited region render for the reflected light effect with all other emitters turned off and that's worked out well. cheers for all your advice.

You should try and combine emitter sliders in this case if you don't necessarily need separate control for each. You can do this by applying the same emitter material to several objects - they will all be represented by one slider in ML.

[lifeofdave]

Is your 'simple chrome reflector' roughness 0 ? - It should be, that will speed up caustics greatly.
A smooth 16bit undulating normal map - mimicing the glazed facade surface which the sunlight is bouncing off.
Rather than use maxwell sun I would use an emitter 500m away + multilight...

Dario said something along those lines about a smooth undulating surface.

Yup, the reflector is roughness zero, and the normal map is 16bit tiff. Using an emitter instead of the maxwell sun is something I need to do more tests with. Most the time I'm setting up the scene in Studio so it's practical just to use the maxwell sun rather than importing a poly in the correct position for a given time of day + date.

Maybe you can make a pass just for this with no other emitters or lights and add it in photoshop to the main pass in linear dodge mode.

Good plan, this adds the possibility of using plain materials and using some software noise reduction as well. Might give this a shot next time.

Here's a screengrab showing part of the final rendered reflection 'pass' at sl 16.6, the light reflected onto the actual building looks great but unfortunately I can't post it here at the moment without client permission.