PDA

View Full Version : Animated Cache Tips? Finding that 'sweet spot'



Matt
06-15-2009, 02:08 PM
I set an animation going yesterday with animated cache on, with what I thought were reasonable GI settings, and left the Preprocess mode to 'Automatic'.

The times started out about 7-8mins per frame, came back from work today and it's taking 1.5hrs per frame!!!!!!!!!

I suspect this is because there are too many samples points being added to the cache as it goes on.

Finding that sweet cache setting is not easy.

Decided to try this 'strategy':


Set cache 'Frame Steps' to 1

Bake entire animation

Set Preprocess mode to 'Locked'

Go to a frame that has a lot going on near the end, do a test with show samples to check how dense it is, if okay render time ...

Render the animation

If not, clear cache, adjust settings and start again


Now I know I run the risk of flickering if there are any frames that have areas that may have been missed, but what do you think, a good strategy?

What are other peoples strategies when it comes to using animated cache and finding that 'magic' setting?

(For the record, I can't use non-interpolated MC, it just takes too long per frame, I've tried).

Discuss ...

:)

Larry_g1s
06-15-2009, 02:20 PM
Tom will be doing a lighting tutorial training that will be acompaning the LWCAD training I'm doing. I believe these are some of the things he'll be covering. If not I'll suggest it to him.

Matt
06-15-2009, 02:32 PM
Exception Tom?

I've clearly read his GI guide, so it's more workflows for finding good settings as oppose to what the settings do, I get that part (just!)

:)

I just wondered how others went about it, especially when encountering the dreaded loooooong later frame times when using animated cache.

Larry_g1s
06-15-2009, 03:14 PM
Yeah, that Tom. I believe sense this will be a paid video training it'll be more then just his on-line write up. More of the "how's & why's". But this might be a good thread for him to view some users concerns or questions.

jwiede
06-15-2009, 03:58 PM
Matt, just out of interest, I presume you don't have FPrime (or it isn't fast enough)?

I'd certainly love to see some training that describes how to identify whether scenes are viable candidates for anim. caching, versus ones which will go into "bazillion additional samples mode". Right now, anim. cache feels too much like a craps shoot where you sink a bunch of time into trying different settings and test renders, but then frequently wind up switching render mode to something else.

I guess what I'm trying to say is that I need greater predictability from it in output times.

Matt
06-15-2009, 05:18 PM
Yes I do have FPrime, but I want to keep this on the topic of LightWave animated caching, tips and techniques for getting a good cache file without too much hassle.

geo_n
06-15-2009, 07:30 PM
I set an animation going yesterday with animated cache on, with what I thought were reasonable GI settings, and left the Preprocess mode to 'Automatic'.

The times started out about 7-8mins per frame, came back from work today and it's taking 1.5hrs per frame!!!!!!!!!

I suspect this is because there are too many samples points being added to the cache as it goes on.

Finding that sweet cache setting is not easy.

Decided to try this 'strategy':


Set cache 'Frame Steps' to 1

Bake entire animation

Set Preprocess mode to 'Locked'

Go to a frame that has a lot going on near the end, do a test with show samples to check how dense it is, if okay render time ...

Render the animation

If not, clear cache, adjust settings and start again


Now I know I run the risk of flickering if there are any frames that have areas that may have been missed, but what do you think, a good strategy?

What are other peoples strategies when it comes to using animated cache and finding that 'magic' setting?

(For the record, I can't use non-interpolated MC, it just takes too long per frame, I've tried).

Discuss ...

:)

In my experience you will always get flicker and dark splotches when set to Locked. Frame steps 1 is prerequisite to AC, too, no success with frame steps every nth frame.
Less lights probably better for AC. Can you post a scene file with proxy objects so its easier to test.

jameswillmott
06-15-2009, 07:54 PM
Yes I do have FPrime, but I want to keep this on the topic of LightWave animated caching, tips and techniques for getting a good cache file without too much hassle.

Can you post an animatic to show us what's being animated?

Matt
06-16-2009, 04:43 AM
Can you post an animatic to show us what's being animated?

It's the robot arm one I'm doing:

http://forums.newtek.com/showthread.php?98204-Mechanical-Arm-WIP&p=893747#post893747

I messed with the settings (A LOT) last night again, which did get further than my previous attempt and seemed to be maintaining a reasonable render time (e.g. it wasn't on 1.5hrs / frame from the original 7mins / frame!)

But I quickly (using QuickTime Pro player's load image sequence) built the rendered frames to see flickering, a big shift in light change, which was unacceptable, so I killed the render and cleared the cache (AGAIN).

Left the changes I made to AA / AT etc. and upped the RPE to 256 (probably overboard) and SBR to 96.

There must be a better way, as it's almost impossible to predict what settings to use that doesn't cause flickering, but also doesn't choke the cache with a bazillion sample points further down the animation, without doing a full render of the whole thing to check.

I'm rapidly coming to the conclusion that Animated Cache is just soooooo picky about having the right settings that it's simply not worth bothering with, might as well bite the bullet and just use slooooooooooooow non-interp MC rendering.

And I was just starting to like LightWave again. :(

JeffrySG
06-16-2009, 08:56 AM
I don't have any info for you, Matt but I'm eagerly following this thread. I'm very curious to read what solution you end up using. I hope there is a good solution whichever way you end up going.

Matt
06-16-2009, 12:49 PM
Same here. I would love the Newtek boys/girls to come in here and explain/provide answers. It would be a BIG help. I won't be ready for animating for about two months, but it would certainly be nice to have a great place to start BEFORE I get to that point.

That's the thing though, I get what the settings are doing and roughly how they work, thanks to Tom (Exception).

I was hoping for peoples thoughts on strategies for getting to that 'magic setting' fast, and without a ton of test renders that never quite work out (i.e. too much wasted time, something I have little of these days) :(

I have my strategy for stills, but animation, especially with reasonably smooth, flickerless GI STILL seems to be the holy grail in LightWave.

Not saying it's impossible, Mark did wonders with the GI, it's just when you want to animate it! It seems to be very frustrating getting there, especially when you try almost everything and are none better of at the end of it.

Matt
06-16-2009, 01:26 PM
I think deforming geometry is another matter, I'm just trying to get something that looks good with 'normal' moving geometry.

pooby
06-16-2009, 01:49 PM
Call me a naive fool, but It would be very straightforward to write each GO frame to a cache, which, then, at the end of the sequence this cache is analysed, to detect generalised true lighting shifts, and eliminate blotchy, one-frame-wonder flashes due to the interpolation.
Then this cache is post processed to smooth it all out and the post processed one is used in the actual render. this would work for deformation and everything.

Thats what I would do if I were in charge.. in fact, I'm sure if you baked the whole scene out to image maps, you could pretty much replicate this procedure in a compositing program.

But yet they faff about with sticking sampling points to geometry and such things.

GraphXs
06-16-2009, 08:03 PM
Yes, why is it so complicated to get a fast GI solution for any situation? Does AC need to look that far between the frames, can the GI rays that fire be locked to the location (like baking) until object postions passes into that ray. Or some type of blending of pixels or rays? What about how game engins do rendering tricks, can some type of real-time tech render engin be part of LW's GI solution? A more low level faked solution?

Looking forward to another fourm about LW GI.

geo_n
06-16-2009, 08:09 PM
It's the robot arm one I'm doing:

http://forums.newtek.com/showthread.php?98204-Mechanical-Arm-WIP&p=893747#post893747

I messed with the settings (A LOT) last night again, which did get further than my previous attempt and seemed to be maintaining a reasonable render time (e.g. it wasn't on 1.5hrs / frame from the original 7mins / frame!)

But I quickly (using QuickTime Pro player's load image sequence) built the rendered frames to see flickering, a big shift in light change, which was unacceptable, so I killed the render and cleared the cache (AGAIN).

Left the changes I made to AA / AT etc. and upped the RPE to 256 (probably overboard) and SBR to 96.

There must be a better way, as it's almost impossible to predict what settings to use that doesn't cause flickering, but also doesn't choke the cache with a bazillion sample points further down the animation, without doing a full render of the whole thing to check.

I'm rapidly coming to the conclusion that Animated Cache is just soooooo picky about having the right settings that it's simply not worth bothering with, might as well bite the bullet and just use slooooooooooooow non-interp MC rendering.

And I was just starting to like LightWave again. :(

Curious why did you go with AC on that situation. Its a single object with no surrounding model. I think BG radiosity is faster and easier which is lw strength up to now.
I think AC will be much more useful for Isolated objects in an enclosed space where BG radiosity can't be used.

Matt
06-16-2009, 08:44 PM
Curious why did you go with AC on that situation. Its a single object with no surrounding model. I think BG radiosity is faster and easier which is lw strength up to now.
I think AC will be much more useful for Isolated objects in an enclosed space where BG radiosity can't be used.

Background radiosity doesn't do multiple bounces. I suppose I could have used that and added area lights to fill in the missing bounces.

But I also wanted to try out animated caching, not had a project where I could use it in anger until now.

Matt
06-17-2009, 02:22 AM
Brilliant.

So I've just had to cancel after two days of rendering on yet another attempt of using Animated Caching.

Has anyone consistently had decent results with this? I just can't get it to work at all, it just seems to be offering either astronomic render times or light blotches for me at the moment.

:(

GraphXs
06-17-2009, 06:23 AM
Can ya just do straitght MC/FC interpolated radosity with high setting?

jameswillmott
06-17-2009, 06:30 AM
Can ya just do straitght MC/FC interpolated radosity with high setting?

Might be faster than the animated cache too... if you have high enough settings you won't notice any flickering.

geo_n
06-17-2009, 06:47 AM
Brilliant.

So I've just had to cancel after two days of rendering on yet another attempt of using Animated Caching.

Has anyone consistently had decent results with this? I just can't get it to work at all, it just seems to be offering either astronomic render times or light blotches for me at the moment.

:(

I think bg rad and area lights is the way to go with a single object. Just cheat the extra bounces:D.
There's no way around long render times in AC.

Matt
06-17-2009, 07:09 AM
I've had to give up on AC at the moment, going to use FPrime, since it's version of brute force MC is waaaaay faster than NewTeks.

Scazzino
06-17-2009, 10:19 AM
Has anyone consistently had decent results with this? I just can't get it to work at all, it just seems to be offering either astronomic render times or light blotches for me at the moment.

For animation I still use spinning lights. When the animated cache first came out I gave it a good couple of days worth experimenting and quickly determined it wasn't worth the effort especially due to the unpredictability of subsequent frame render times. Since I need motion blur and deformations, I need to use classic motion blur passes anyway, which makes spinning lights basically free since they are calculated at the same time...

Panikos
06-18-2009, 12:44 AM
Chasing the sweet spot, may take forever. Its something very sensitive and depends on scene content.

Matt
06-18-2009, 03:07 AM
Chasing the sweet spot, may take forever. Its something very sensitive and depends on scene content.

True, but when something is like trying to get cats in a bag, I simply won't ever use it, because on a job with a deadline I simply don't have the time to mess about with it.

Software is meant to make your life easier / do things faster, right?

Just to add, I wasn't after 'one size fits all' settings with this thread, it was _how_ people go about finding the right settings and what strategies they employ to do so.

Panikos
06-18-2009, 03:33 AM
Having said that, a "Rendering Presets" is not something effective

Matt
06-18-2009, 04:41 AM
Having said that, a "Rendering Presets" is not something effective

Not presets with caching no.

Paul Brunson
06-19-2009, 06:08 PM
I had the trouble with the animated cache as you described. It bogged down and became crazy slow with each passing frame. I then set it to every 4th frame and the render times became much more reasonable with no flicker that I can see.

Here is a incomplete test animation made with the every 4th frame for the cache baking. I checked my butterfly rendering logs and the render time stayed steady between 15 to 20 minutes a frame through all 800 frames.


http://www.3dmusings.com/CacheTestRender.mov
(Note: It appears I made a bad choice of compression, there is some flickering in this that's definitely not in the uncompressed rendered frames...will try another codec and post again)

The scene also has 3 area lights to create some soft but more defined shadows. Its surrounded with a luminous white sphere. My GI settings for the scene are as follows:

Intensity: 70%
Indirect Bounces: 3
RPE: 300
SBR: 60
Angular Tolerance: 10.0
Min Pixel Spacing: 1.0
Max Pixel Spacing: 100
Multiplier: 100%

Cache: Locked
Cache Frame Step: 4

Paul Brunson
06-19-2009, 06:29 PM
Ok... here is a better version.

http://www.3dmusings.com/CacheTestRender_v2.mov

Matt
06-19-2009, 07:12 PM
You see that looks okay, maybe AC has problems with fast moving objects? My animation moves pretty quick (deliberately so).

geo_n
06-19-2009, 08:18 PM
Ok... here is a better version.

http://www.3dmusings.com/CacheTestRender_v2.mov

This animation doesn't need AC imho. It can be done faster and easier with backdrop radiosity with area lights like pre lw 9 way.

Paul Brunson
06-20-2009, 01:21 AM
quite possibly so, I found myself liking the effect of the bounce rays tinting things just slightly. I haven't really looked at optimizing the render times yet, at which point I'll have to decided if the subtle bounce light is worth it.

I think that point aside though, the important thing is that the render times per frame remain steady rather than growing over time while using animated cache.

geo_n
06-20-2009, 02:29 AM
quite possibly so, I found myself liking the effect of the bounce rays tinting things just slightly. I haven't really looked at optimizing the render times yet, at which point I'll have to decided if the subtle bounce light is worth it.

I think that point aside though, the important thing is that the render times per frame remain steady rather than growing over time while using animated cache.

Do you mind posting the scene file? I'd like to test render how fast it is in AC and then do another render with bg rad.
Can you speed up your animation and still keep frame step to 4 and locked? I haven't had good results with faster moving objects not set to everyframe and automatic without using motion blur. But motion blur in 3d is slow so tendency not to use iti.

Paul Brunson
06-20-2009, 12:20 PM
Unfortunately the objects in that scene I'm not able to distribute. However I stayed up way too late last night creating a test scene and doing test renders. My hope is to specifically test fast moving objects (with motion blur) using the AC (animated cache). I'll gladly post that with results when done.

Intuition
06-22-2009, 02:59 PM
My buddy Stefan Brederock and I found a great way to do GI in Lightwave.

I was always afraid of posting it since I feared it was a bug and would be "corrected" in future releases.

When calculating GI in LW one often turns up the rays and samples such that the GI will look good in a still frame with decently high settings.

It turns out that using animated cache is trying to compensate for the different samples in each frame by looking forward.

This method barely works, unless you know specific voodoo magic.

I sidestep this method by utilizing the GI the way other engines do it.

Use motionblur passes to sample the GI with very low sampling so that when all the passes are added up together for the final frame the actual blend makes the GI solution very smooth, clean and accurate. Since it is blending it will also make a smooth AA solution even when you use low AA samples.

A person told me this method wasn't a good idea in the beta forums in 9.3 . Well it actually works very well.

You will find you can even do this method with straight monte carlo.

You need to use perspective camera.

What you will do is take whatever scene you are trying to light up and go to the camera properties and turn down the AA samples to like 2 or 3. Also, don't use Adaptive sampling or over sampling at first.

Now go to the GI panel and use final gather interpolated (though this works in MC as well) and use like 24 rays and 12 secondary rays. at angular tolerance 45 degrees and do a min pixel spacing of 1 with a maximum of say 30. The min max can be set to what ever you need but for this test just use these numbers.

Now go back to your camera properties and turn on dithered motion blur to like 24 passes. Whah, you say?

Yup, Dithered motion blur. For som reason this motion blur method will sample every pass and blend them together at the end. This is great because you can now fire off low sample settings in your GI and AA samples and the overll effect of 24 passes of dithered motion blur will blur them together to such an extent that it works like the old spinning spotlight trick except we are now doing it for the AA and the GI at the same time.

The result will be a motion blur that looks like photreal motion blur and a GI that is much less expensive but looks as good as an hour+ a frame GI.

Now I had frames between 2-5 minutes a frame at 800x600 for some serious geometry w/GI that looked pretty damn good. It was still flickering but very very very slight...so.... I turned up the settings a little more and ended up with 12 minute a frame GI that was flawless.

I'll post pics and a scene file when I get home. This really is the only way to fly with GI in LW.

I hope that it won't get fixed and perhaps Newtek can look at this method and apply it to work similarly with PR motion blur.

I have used it with straight monti carlo set to like 12 samples with only 2 AA samples in the camera and then had like 30-40 dithered passes and the frame renders pretty fast and looks better then a frame set to 256 passes which renders much much slower.

Credit goes to Brederock at EdenFX for showing me that dithered Mblur was working how I was trying to get PRmblur to work. This has been a little secret of ours since 9.2 - 9.3 beta.

Panikos
06-22-2009, 03:12 PM
Thats "distributed raytracing" it was invented by Cook in 1986 if I am not mistaken.
The principle aplies for GI too.

Paul Brunson
06-22-2009, 09:24 PM
Great minds think a like, I've been fiddling with a method that uses a similar idea. (Yours sounds like it might be even faster though :)

I got the idea after watching Fprime render a scene and reading Exceptions article on the new Adaptive sampling antialiasing that we got with LW 9.3.

Rather than launch into a long explanation I'll post the proof. Attached are two images. One rendered in Lightwave, the other in the fprime. The Fprime image was given the same amount of time to render as the Lightwave frame took to render. As you can see the Lightwave image is smoother and grain free.

Also see this animation: http://www.3dmusings.com/roomtest_v001.mov
(I suggest saving it to disc and looping the animation as its short.)

The images and animation were rendered at 960x540 (50% of full HD 1920x1080)

On my quad core 2.4 ghz AMD Opteron machine it took about 8 minutes a frame with 8 bounces of Radiosity and photoreal motion blur . I imagine on these new Fandangle Nehalem based chips it would render in about 4 minutes or so.

Intuition
06-22-2009, 09:56 PM
Here is a small test scene showing the idea. Its basic settings show how you can send a sequence of low sampling settings (AA and GI) into a multipass dither MBlur and get a nice render in the end at low frame times.

Now you can expand on this concept by switching to higher settings in the GI (more GI samples primary and secondary) or add even more dither mblur passes.

Usually I can even use interpolated radiosity with this method though it takes 30-50 passes to fully clean up flickering yet you will still see its much much faster then other methods. Also, try switching the settings to straight Monti Carlo as well. ;)

geo_n
06-22-2009, 09:58 PM
My buddy Stefan Brederock and I found a great way to do GI in Lightwave.

I was always afraid of posting it since I feared it was a bug and would be "corrected" in future releases.

When calculating GI in LW one often turns up the rays and samples such that the GI will look good in a still frame with decently high settings.

It turns out that using animated cache is trying to compensate for the different samples in each frame by looking forward.

This method barely works, unless you know specific voodoo magic.

I sidestep this method by utilizing the GI the way other engines do it.

Use motionblur passes to sample the GI with very low sampling so that when all the passes are added up together for the final frame the actual blend makes the GI solution very smooth, clean and accurate. Since it is blending it will also make a smooth AA solution even when you use low AA samples.

A person told me this method wasn't a good idea in the beta forums in 9.3 . Well it actually works very well.

You will find you can even do this method with straight monte carlo.

You need to use perspective camera.

What you will do is take whatever scene you are trying to light up and go to the camera properties and turn down the AA samples to like 2 or 3. Also, don't use Adaptive sampling or over sampling at first.

Now go to the GI panel and use final gather interpolated (though this works in MC as well) and use like 24 rays and 12 secondary rays. at angular tolerance 45 degrees and do a min pixel spacing of 1 with a maximum of say 30. The min max can be set to what ever you need but for this test just use these numbers.

Now go back to your camera properties and turn on dithered motion blur to like 24 passes. Whah, you say?

Yup, Dithered motion blur. For som reason this motion blur method will sample every pass and blend them together at the end. This is great because you can now fire off low sample settings in your GI and AA samples and the overll effect of 24 passes of dithered motion blur will blur them together to such an extent that it works like the old spinning spotlight trick except we are now doing it for the AA and the GI at the same time.

The result will be a motion blur that looks like photreal motion blur and a GI that is much less expensive but looks as good as an hour+ a frame GI.

Now I had frames between 2-5 minutes a frame at 800x600 for some serious geometry w/GI that looked pretty damn good. It was still flickering but very very very slight...so.... I turned up the settings a little more and ended up with 12 minute a frame GI that was flawless.

I'll post pics and a scene file when I get home. This really is the only way to fly with GI in LW.

I hope that it won't get fixed and perhaps Newtek can look at this method and apply it to work similarly with PR motion blur.

I have used it with straight monti carlo set to like 12 samples with only 2 AA samples in the camera and then had like 30-40 dithered passes and the frame renders pretty fast and looks better then a frame set to 256 passes which renders much much slower.

Credit goes to Brederock at EdenFX for showing me that dithered Mblur was working how I was trying to get PRmblur to work. This has been a little secret of ours since 9.2 - 9.3 beta.



Cool tip. Can work with deformations and moving camera.

Intuition
06-22-2009, 10:00 PM
Yes it works especially well with moving camera and deformations.

---had to fix the dload link zip file. See if it works now. :-/

Oh. Also the settings in the scene file are interpolated, try turning off the interpolated as well with both MC and FG. You'll see some cool results.

geo_n
06-22-2009, 10:03 PM
Great minds think a like, I've been fiddling with a method that uses a similar idea. (Yours sounds like it might be even faster though :)

I got the idea after watching Fprime render a scene and reading Exceptions article on the new Adaptive sampling antialiasing that we got with LW 9.3.

Rather than launch into a long explanation I'll post the proof. Attached are two images. One rendered in Lightwave, the other in the fprime. The Fprime image was given the same amount of time to render as the Lightwave frame took to render. As you can see the Lightwave image is smoother and grain free.

Also see this animation: http://www.3dmusings.com/roomtest_v001.mov
(I suggest saving it to disc and looping the animation as its short.)

The images and animation were rendered at 960x540 (50% of full HD 1920x1080)

On my quad core 2.4 ghz AMD Opteron machine it took about 8 minutes a frame with 8 bounces of Radiosity and photoreal motion blur . I imagine on these new Fandangle Nehalem based chips it would render in about 4 minutes or so.

Can you try more than 1 light source and a non fixed camera? I'm not exactly familiar with your pc spec but 8 min seems a lot 2.4ghz quad for that render. I think the motion blur is causing big render time but yeah its one way to smoothen renders even before.

Intuition
06-22-2009, 10:05 PM
Grrr the settings I uploaded were too high I just checked the file....grrr. One sec, A new file is coming. I was trying to start with 8 over 4 and it ended up 64 over 32.

Intuition
06-22-2009, 10:08 PM
Ok start here. This is a good basis.

Its 8 over 4 FG @ 20 dither passes. I do render on an 8 core so yeah quad cores are 2x my render times. Balance the render times for your computer, try 12 dither passes at first. Keep going up with passes or samples till you find a good balance.

Intuition
06-22-2009, 10:20 PM
You can see that having this nice of a render at like 1-2 minutes will make it so that you can tune it so its flawless at 12-15 minutes a frame which just kills if you have a big render farm as opposed to 1+ hour frames which just end up still flickering.

Paul Brunson
06-22-2009, 10:30 PM
Can you try more than 1 light source and a non fixed camera? I'm not exactly familiar with your pc spec but 8 min seems a lot 2.4ghz quad for that render. I think the motion blur is causing big render time but yeah its one way to smoothen renders even before.

My machine was cutting edge about 3-4 years ago. Like I say these days a quad-core intel chip at the same ghz would render it in about 4 minutes.

Machine speeds vary, it can be a beast to compare render times with any sort of accuracy. Which is why I posted the Fprime render as a comparison, as everyone is always saying Fprimes radiosity is so much faster than Lightwave's.

I'll bundle up the scene and let you take a whack at speeding it up if you like.

Paul Brunson
06-22-2009, 10:44 PM
Just gave your v2 scene a try Intuition, looks very promising, going to try to apply your technique to the scene I've been fiddling with up above.

(Just as a reference my quad core 2.4 ghz AMD opteron took 4m44s to render Intuitions "GI-Box-Area_v02" scene with no alterations.)

Paul Brunson
06-22-2009, 11:09 PM
Intuition your technique is fantastic! Yields a better result than my previous tests in less than half the time! (3m18s)

I was trying to get the Adaptive Sampling system to do a similar thing, layer noisy GI on top of noisy GI to get a smooth solution. But this does it so much better its not even funny. (If fact its sad, all that time I've wasted trying other techniques.)

See below for an image from Lightwave with Intuitions Technique, and just for fun I did an Fprime render for the same 3m18s. There's not even a comparison, Fprime still has way too much grain.

Panikos
06-22-2009, 11:45 PM
See below for an image from Lightwave with Intuitions Technique, and just for fun I did an Fprime render for the same 3m18s. There's not even a comparison, Fprime still has way too much grain.

Thats very natural.
LW blends several "clean" or inacurate pictures to get a steady result.
FPrime composes several noisy but accurate pictures to get a steady result.

If we try both ways at an extreme test, FPrime will be more reliable, believe me.

Paul Brunson
06-23-2009, 12:00 AM
I would be interested to see a case where fprime accuracy beats out this technique. For me personally even before this technique I used fprime as previewer only and rendered my finals with Lightwave because I've always been able to get smoother results quicker.

Fprime renders fast initially but then takes for ever to kill the grain and get a smooth result in my experience.

But I would love to be proven wrong as I'd love to see my investment into the plugin get some more mileage.

Panikos
06-23-2009, 12:09 AM
There are no secrets in FPrime. Everything is written in the online manual, how to kill noise.

FPrime is more reliable as I wrote before, because its samples are not random but satisfy a specific order.

Lets take the roulette for example.
LW put bets in many places
FPrime put bets in all places one at a time.
Both WIN!!!

However FPrime does not miss a chance.

If you understand this, you know why FPrime is more reliable cause it accounts everything during its calculations.

Intuition
06-23-2009, 01:19 AM
Panikos is right about f-prime.

There is a bell curve of scene complexity where this technique will fall off and F-Prime will pull ahead.

For most stuff this will work really well but there are some levels of complexity where LW will just slow down, no matter the method used, and F-Prime or Kray will be much much better but, I figured that based on the pictures in the gallery that this technique can be plenty useful for most users. :D

I mean, once you start to slow down with 1.0 minimum pixel spacing you can get some more leverage with 2.0-4.0 minimum pixel spacing to save some time again. Not as accurate buts till saves time on more complex scenes and can be cleaned up with passes. None the less. Kray or F-Prime can be better at such complex scenes like greeble masterpieces and such.

Panikos
06-23-2009, 01:21 AM
Megalodon
I am glad that you are happy.
I dont get commision out of anybody's sales.
You are free to interpret things as you like :)

toby
06-23-2009, 04:59 AM
The motion blur method works ok, but if you compare it to a render with higher settings (I usually end up with about 96) you'll see a definite difference. Also, it can take longer if your scene doesn't need all the passes for good motion blur. I rendered this scene 4 times faster with rays at 96 and good motion blur, at 8 passes - but if things were moving faster (like any vehicle going past a static camera) you'd see steps in the motion blur. So it seems like you should determine how many passes will be needed for a given shot/scene, then set the gi samples accordingly.

Intuition
06-23-2009, 11:24 AM
Yeah there are certain cases where you have to balance the MB passes vs the GI samples. In some scenes lower GI and AA will get smoothed out by dither passes much faster then setting AA to like 5-7 and GI 48-96+ primary samples.

Its not an end all solution for all scenes but I've seen it make great frame times over and over that were just much better then high samples with prmblur passes.

In the end it shows that there is plenty of optimization room left in LW's native engine. It makes me feel that PRmblur could be setup to work the same way if could sample GI over passes like dithered does.

Some fancy coding ideas and it could be nice.

Intuition
06-23-2009, 11:25 AM
Yeah there are certain cases where you have to balance the MB passes vs the GI samples. In some scenes lower GI and AA will get smoothed out by dither passes much faster then setting AA to like 5-7 and GI 48-96+ primary samples.

Its not an end all solution for all scenes but I've seen it make great frame times over and over that were just much better then high samples with prmblur passes.

In the end it shows that there is plenty of optimization room left in LW's native engine. It makes me feel that PRmblur could be setup to work the same way if could sample GI over passes like dithered does.

Some fancy coding ideas and it could be nice.

Now days I render most stuff in Vray anyways, which just kills anything I throw at it, but this trick helps with LW centric renders when I use LW.

toby
06-23-2009, 03:29 PM
Yea that's the tricky part, you'd have to judge where the fastest motion is and, probably starting with high settings, lower it until you see steps, then raise it back up a little. You do risk missing something unexpected, which is why at places like dd you tend to use many dozens of passes regardless, so Intuitions method works right of the bat.

adk
06-23-2009, 09:16 PM
I've just been playing around with the test scene and techniques that were posted & wanted to say cheers a bunch to you all for a wealth of information on this elusive topic :thumbsup:

Maybe I'm doing something wrong / or right ... or misunderstanding something quite crucial but I get way faster and smoother results than those posted by Intuition & toby ... and I'm using the animated cache.
Quality is ... well you be the judge.

Bear in mind these were done on an old battleaxe I have here at work so I had to scale the render down to 50% to get reasonable time frames. Times given are for the whole sequence to render.

...also I turned on raytrace reflections on toby's scene to match the others.

Test 01: 443 mins - Intuition
Test 02: 130 mins - toby
Test 03: 75 mins - me
Test 04: 32 mins - me

On both 01 & 02 there's some slight but noticeable flickering, 03 & 04 seem to be much better. With 03 & 04 I was trying to see if I could get better contact shadows & overall results.



adk.

Paul Brunson
06-23-2009, 10:14 PM
Adk, would you mind sharing your versions of the scene. I'd like to give it a try and compare.

geo_n
06-23-2009, 10:24 PM
I've just been playing around with the test scene and techniques that were posted & wanted to say cheers a bunch to you all for a wealth of information on this elusive topic :thumbsup:

Maybe I'm doing something wrong / or right ... or misunderstanding something quite crucial but I get way faster and smoother results than those posted by Intuition & toby ... and I'm using the animated cache.
Quality is ... well you be the judge.

Bear in mind these were done on an old battleaxe I have here at work so I had to scale the render down to 50% to get reasonable time frames. Times given are for the whole sequence to render.

...also I turned on raytrace reflections on toby's scene to match the others.

Test 01: 443 mins - Intuition
Test 02: 130 mins - toby
Test 03: 75 mins - me
Test 04: 32 mins - me

On both 01 & 02 there's some slight but noticeable flickering, 03 & 04 seem to be much better. With 03 & 04 I was trying to see if I could get better contact shadows & overall results.



adk.

test 01 has solid gi but with some noise. test 02 has slightly warping wall at the end but less noise. test 03 and test 04 has warping wall ang blinking colors. good tests.

adk
06-23-2009, 11:11 PM
adk, how will your setup work for a figure where its mesh deforms?

Not too well obviously Megalodon (I did a quick test so check the .mov attached below - tho that can probably be improved upon plus you have meshes intersecting here which also doesn't help I'm sure ). I'm afraid I was taking Matt's original post into account re. robot arm & render speed and this seems pretty quick and smooth in this case.

As always it seems to be "horses for courses" & nothing works perfect for all occasions.


Adk, would you mind sharing your versions of the scene. I'd like to give it a try and compare.

I'll post 'em all up in a tick. Still at work and have few things to finish up.
Nothing revolutionary I'm afraid ... simply cached interpolated MC.

adk

Panikos
06-23-2009, 11:26 PM
Immediately after the LW9.6 release, my attention was focused on latest GI.
I made a Cornell Box like together with a character that was displaced with MDD.
I tweaked the GI settings for best results and rendered using Animated cache.
As long as the rendering wasnt interrupted the results were excellent.
When I stopped the computer and continue later, there was a sharp difference in the GI solution. This happened on the same computer.
If I had this test made on a Network, I guess all rendered frames would vary.

So, with sorrow I concluded that current LW GI, is not mature enough for animated displacements.

Intuition
06-24-2009, 12:15 AM
I've just been playing around with the test scene and techniques that were posted & wanted to say cheers a bunch to you all for a wealth of information on this elusive topic :thumbsup:

Maybe I'm doing something wrong / or right ... or misunderstanding something quite crucial but I get way faster and smoother results than those posted by Intuition & toby ... and I'm using the animated cache.
Quality is ... well you be the judge.

Bear in mind these were done on an old battleaxe I have here at work so I had to scale the render down to 50% to get reasonable time frames. Times given are for the whole sequence to render.

...also I turned on raytrace reflections on toby's scene to match the others.

Test 01: 443 mins - Intuition




adk.

Wow, 443 mins? Can't say I've seen many renders in LW take that long, even old school methods. Is it a single core computer?

Intuition
06-24-2009, 01:09 AM
I did a quick greeb cubes box test with this method.

I had 1 min 8 secs on the non-reflective ones and 11 minutes in the reflective ones and the reflection/recursion rays are only set to 4. 8 core system 2.5ghz. LW has trouble with reflections still.

adk
06-24-2009, 01:17 AM
Wow, 443 mins? Can't say I've seen many renders in LW take that long, even old school methods. Is it a single core computer?

Heya Intuition ... firstly many many thanks for sharing your setup & technique. I've learned a lot over the years from you kind folks here in this perpetual LW school of ours.

It's a six year old HP 6200 with a dual Xeon so yeah it's pretty old in the tooth but that's for the whole sequence of frames. It was about 4 mins per frame @ 1/2 scale.

Here's another deformation test that's a lot smoother than the last. Not 100% I know but a whole lot better than the previous one I posted. Took 35 mins to render the whole sequence.


adk

toby
06-24-2009, 02:32 AM
Maybe I'm doing something wrong / or right ... or misunderstanding something quite crucial but I get way faster and smoother results than those posted by Intuition & toby ... and I'm using the animated cache.
Quality is ... well you be the judge.
No you're not doing anything wrong, me and intuition just never optimized the gi for this scene. But it looks like the area light quality is lower? I am (pleasantly) surprised that it went faster with the anim cache on!

JonW
06-24-2009, 02:56 AM
I am getting better & quicker results with GI but still playing around with it for architectural scenes, even if I had 30 minutes a frame, this would be ok if I get consistent results.

Frame 1 timings (original):
5450 V8 3.0 Ghz 1:42
940 @ 3.657 Ghz 1:46
920 @ 3.150 Ghz 2:03

geo_n
06-27-2009, 02:13 PM
Ok start here. This is a good basis.

Its 8 over 4 FG @ 20 dither passes. I do render on an 8 core so yeah quad cores are 2x my render times. Balance the render times for your computer, try 12 dither passes at first. Keep going up with passes or samples till you find a good balance.

Just looked at the scene. AC is off?
Should I bake AC with the saved scene settings then after baking turn up all AA, AS etc to my preference?

Matt
06-27-2009, 05:18 PM
Really cool tip Intuition! Thanks for that! :)

Is Mark aware of this tip? I'm wondering if there's anything in it that he could package up as a 'proper' feature, if that makes sense!

Using your settings in your post (forgot to set Adaptive Sampling though!) got me this in 30mins 3s, I'm sure that could be optimised _a lot_ though!

I take it this method works fine for moving lights / objects if you use straight MC and not FG?

geo_n
06-27-2009, 10:18 PM
Really cool tip Intuition! Thanks for that! :)

Is Mark aware of this tip? I'm wondering if there's anything in it that he could package up as a 'proper' feature, if that makes sense!

Using your settings in your post (forgot to set Adaptive Sampling though!) got me this in 30mins 3s, I'm sure that could be optimised _a lot_ though!

I take it this method works fine for moving lights / objects if you use straight MC and not FG?

I'm getting huge render times after the initial frames with Intutions file when I turn on AC finalgather interp>bake>AC locked>turn on AS .05. Just as it would when AC is on. What did you do?

I made a test. 2min render per frame qc6600

Intuition
06-28-2009, 09:02 AM
This method isn't meant to be used with animated cache.

Matt, I have used this method with interp FG as well, though you'll want to keep the GI samples down so you can add more dither passes quickly so you can reduce the flickering. I would like to see if Mark could make LW's general GI better by looking at a new solution that samples over MBlur passes even with photoreal Mblur.

On some scenes in BSG I had to go as high as 48 primary 24 secondary with like 24-48 dither passes. Usually I try to stay lower then that.

I often have to balance things. I try to keep AA at 2 and sometimes I'll put it at 3 if I have problem spots but after you get 24'ish dither passes it gets long. The idea is to let the AA be 2 (no adaptive or oversampling it will spike) and the GI stay low, though I often use 1 min / 30 max pixel spacing I have taken the min pixel spacing up to 4 just to speed things up in certain scenes. At which point I have to take the dither passes up above 24 passes so that the GI flicker will get reduced from the blending.

I usually always can find a soluton though. Scenes like Matt's are primary candidates for this method.

One thing I've noticed is that LW is having a hard time with reflections in this build. Seems like its slower then older LW versions with simple reflections. Maybe causing a bottle neck.

Intuition
06-28-2009, 09:05 AM
I looked at your test Geo and it looks like it has no Mblur. Are you using Dithered Mblur with perspective cam?

-----EDIT=----

Ok I see it. Sorry. ;) It has mblur.

Maybe the Mblur test scene needs to be higher then 50%

geo_n
06-28-2009, 08:24 PM
Ah I see its an old school trick. I thought its Animated Cache tip.:confused:
If using old school trick I think 9.3 is faster. AA and AS is faster in 9.3. I really hoped it was for AC.

That render is not lightwave. Its impossible for lw to get that quality in 2min render time with my machine. :D
I will try toby's file, it might be faster but quality is still not as good looking at adks video.

geo_n
06-28-2009, 09:42 PM
Maybe I'm doing something wrong / or right ... or misunderstanding something quite crucial but I get way faster and smoother results than those posted by Intuition & toby ... and I'm using the animated cache.
Quality is ... well you be the judge.


Can you share the file where you used AC and its faster than Intuition and tobys file? When I turn AC on the next frames are super slow rendering. I don't know how you made it fast on yours:thumbsup:.
Quality wise the videos you posted 03 and 04 have inaccurate GI. Warping walls,etc. Intuition and tobys 01 and 02 video has better GI but some noise.

adk
06-28-2009, 10:41 PM
... sorry folks I got a bit busy since those tests & never made time to UL the settings nor scenes. I'll dig 'em up and post in the next couple of hrs.

All my frames took pretty much exactly the same time ... so maybe I did something incorrect / weird. I'm no expert in this field & just wanted to simply play around with the scenes & settings.

Yeah there is a bit of colour flickering on my tests & the vertical lines warped a bit in places but hey they rendered pretty quickly & gave almost identical results.

adk

adk
06-28-2009, 11:34 PM
... here's the faster of the 2 settings.

It's simply cached interpolated MC with the settings you can see in the pic.

Baking the cache ( at frame step of 2) took about 55 secs. Renders took about 13-15 secs per frame. Lighting setup / quality is exactly the same as the original scene.

I even tried what Panikos mentioned, as in restarted the computer, resumed rendering & checked for GI flickering/anomalies and to my eyes did not get any noticeable changes compared to just doing it in one hit. It could be the size I'm rendering at that's simply smoothing out the errors so it would be good to see what other people get.

Maybe my eyes are just getting old or I'm simply extremely unobjective :) :D

adk

geo_n
06-28-2009, 11:50 PM
... here's the faster of the 2 settings.


Thanks adk. This will help in finding good solution to lw gi render.
Ok so I tried another. This is fast but worse quality than test00.mov.
Render time is 1:20 min per frame. Its not the same renderer as my test00.mov btw.

toby
06-29-2009, 12:24 AM
... here's the faster of the 2 settings.

It's simply cached interpolated MC with the settings you can see in the pic.

Baking the cache ( at frame step of 2) took about 55 secs. Renders took about 13-15 secs per frame. Lighting setup / quality is exactly the same as the original scene.

I even tried what Panikos mentioned, as in restarted the computer, resumed rendering & checked for GI flickering/anomalies and to my eyes did not get any noticeable changes compared to just doing it in one hit. It could be the size I'm rendering at that's simply smoothing out the errors so it would be good to see what other people get.

Maybe my eyes are just getting old or I'm simply extremely unobjective :) :D

adk
Wow those gi settings are very low-res, and you have no aa, it's no wonder it went faster! Min. pixel spacing of 4 will give you clearly inaccurate lighting, and 2 pass motion blur with no aa will look bad too, when viewed at full res.


That render is not lightwave. Its impossible for lw to get that quality in 2min render time with my machine.
Very nice, Kray or Vray? There are some interesting differences, like the motion blur, it's half as long as it is in the LW scenes; it will look just as good while using faster settings.

Another thing I just noticed is that the area light is making the scene take 4 or 5 times longer!!! So, even if other renderers use area lights too, this is *not* an accurate test of gi!

So does the Vray render use an area light geo_n? If not, lw's speed might be more similar.

Another difference I noticed is that it takes 4 bounces to match the amount of color bleed in the Vray render. If the vray render is only 2 bounce, then this is probably just a difference in the way the programmers planned the renderer. If any of you missed 3d4e's tutorial, he explains how to decrease the bleed even more, for a more realistic render. Would be nice to have a slider in the gi panel for that.

geo_n
06-29-2009, 12:58 AM
Wow those gi settings are very low-res, and you have no aa, it's no wonder it went faster! Min. pixel spacing of 4 will give you clearly inaccurate lighting, and 2 pass motion blur with no aa will look bad too, when viewed at full res.


Very nice, Kray or Vray? There are some interesting differences, like the motion blur, it's half as long as it is in the LW scenes; it will look just as good while using faster settings.

Another thing I just noticed is that the area light is making the scene take 4 or 5 times longer!!! So, even if other renderers use area lights too, this is *not* an accurate test of gi!

So does the Vray render use an area light geo_n? If not, lw's speed might be more similar.

Another difference I noticed is that it takes 4 bounces to match the amount of color bleed in the Vray render. If the vray render is only 2 bounce, then this is probably just a difference in the way the programmers planned the renderer. If any of you missed 3d4e's tutorial, he explains how to decrease the bleed even more, for a more realistic render. Would be nice to have a slider in the gi panel for that.

I used both vray and kray. Similar render time results in the 1-2 min range for my qc6600 campared to Intutions file is 5min per frame(but its a really great trick for non AC in lw).
For both kray and vray scene I adjusted the motionblur. I made it less but the rendertimes were the same even with more motionblur. In the vray scene I used the same method as Intuition. Making the white poly vraylightmtl and added 1 vray light(32subdiv) to mimick Lw area light. A little adjustment of colorbleed because there's excessive color bleed in vray and exposure color mapping. In kray I just used interpolated extinction 3% I think with medium presets and fsa for motion blur dithered 15 instead. kray is really fast and I hope update is out for core asap and have full access to core. lw 9.6 is the end of its cycle and I think kray is branching out for other 3dappz. Give vray a good competition.

adk
06-29-2009, 01:12 AM
Wow those gi settings are very low-res, and you have no aa, it's no wonder it went faster! Min. pixel spacing of 4 will give you clearly inaccurate lighting, and 2 pass motion blur with no aa will look bad too, when viewed at full res.


.. my bad on those aa settings posted toby, they should have been set to 2 instead of 1 (render times will be almost exactly the same - I just checked). I kept mucking about with settings and did not save the files I should have.

Settings wise - I know these are very low but I was simply going for a faster and comparable render (if there is such a thing).

adk

toby
06-29-2009, 02:40 AM
I'm starting to think that the motion blur method might be *better* than using the anim cache in the majority of situations. The anim cache wants to freeze where your final gather points are, so if it's blotchy it will keep the same blotchiness throughout the anim - the motion blur method is less blotchy and also renders faster.

6 pass motion blur render at 2min. 43sec.
http://i695.photobucket.com/albums/vv317/schmoby/mblur.png

2 pass moblur, anim cache with the whole anim pre-baked, 4min., same rpe
http://i695.photobucket.com/albums/vv317/schmoby/blotch.png

As usual I'm sure it depends on the scene, and Exception might have something to point out, but the difference in time and quality is significant.

This is done without the area light btw, gi intensity turned up to 400%.

* my mistake, it renders faster with the anim cache not the motion blur, 1min 34sec.

toby
06-29-2009, 02:53 AM
Ok here's the anim cache with more rays, 256 over 128 as opposed to 96 over 48 - still a little slower, 3min 33 sec, still a little blotchier
http://i695.photobucket.com/albums/vv317/schmoby/blotch2.png

Panikos
06-29-2009, 03:38 AM
In areas with insufficient light, the possibility of flickering is bigger cause the range within the calculations take place is limited. Opposed to this, in areas where there is sufficient light, the range of calculations is wider and flickering most likely wont show up.

geo_n
06-29-2009, 04:34 AM
This is your scene file toby. But i changed ray recursion to 8 instead of 1, turned on directional rays and raytrace reflection, refraction. Everything else is the same. The wall seems to be warping. I dont know what is causing that. Flicker on corners. But render time in qc6600 is 1min20 sec. Good speed and acceptable quality for archiviz.
I will try Intuition file and render sequence. So far looks like its the most stable gi but with noise and longest render time.

Exception
06-29-2009, 10:44 AM
Ha, only found this thread now. And what a great one it is!
The method Intuition suggest was the way we used GI in animation years and years ago. I nearly forgot about it. And a great solution it is. Thanks for bringing this back into the daylight KC.

That said, I tested Matt's animation extensively a week or so back, and really could not reproduce the trouble he had with it. I had no slowdowns at all, no matter what I did. Problem is, with AC, I feel, that it's really very immature currently. These freak occurences which might totally wreck an intended project setup occur a bit too frequent, and then make you look like an idiot trying to find the bug or the cause.

The system of an independent cache per frame and interpolate/blend over a certain frame range is the system Kray uses. I've suggested this method, alternatives to it and others several times to the Dev team. The Kray method leaves something to be desired, for sure, but there's other wasy to handle it that would possibly work better. That said, I'm pretty sure they're not easily programmable, and I do know Mark has known of a way to make Animated GI caching work with deformations, but that the old LW code restrictions kept him from leveraging it. So it might be something that will only surface in CORE. (sigh)

Intuition
06-29-2009, 01:57 PM
Just in case anybody missed details in my post, you CAN NOT USE PHOTOREAL MOTIONBLUR to make this work.

YOU MUST USE DITHERED MOTION BLUR TO MAKE THIS METHOD WORK.

It is the only mblur method that blends the many motion blur passes GI samples together as one. Photoreal motion blur will not work you will have blotchies.

You may still get blotchies with FG interpolated / dithered but you can always give it enough passes to smooth it out. The trick is to balance the samples min max and the primary secondary so that they are low enough to keep the individual motion blur passes (Dithered a/b) fast to render, which is why I don't use AS at all. The Dithered Mblur will smooth out the AA after 16+ passes out of pure pass blending.

toby
06-29-2009, 03:26 PM
This is your scene file toby. But i changed ray recursion to 8 instead of 1, turned on directional rays and raytrace reflection, refraction. Everything else is the same. The wall seems to be warping. I dont know what is causing that. Flicker on corners. But render time in qc6600 is 1min20 sec. Good speed and acceptable quality for archiviz.
Which scene, the first one I posted? I have a slow machine too, so I always turn off non-essentials in all scenes that I'm comparing. Faster test renders allows more tests for greater accuracy!

Which wall seems to be warping? I see some flickering, so it does need more passes, but I'm not sure what/where you're refering to.

Cageman
06-29-2009, 04:14 PM
I set an animation going yesterday with animated cache on, with what I thought were reasonable GI settings, and left the Preprocess mode to 'Automatic'.

The times started out about 7-8mins per frame, came back from work today and it's taking 1.5hrs per frame!!!!!!!!!
:)

That is wierd... usually, only the first frame is the one that can take alot of time, any subsequent frames should go faster... at least that is what I've experienced so far....

geo_n
06-29-2009, 04:41 PM
toby - this scene Gi-Box-Area_tg_V03.lws.zip
the walls have subtle errors when objects are in front while moving. very slight.

This is Intutions scene I just renderer a few frames. I just turn on interpolated FG and there's splotchiness on the shadows. Did I make mistake?

geo_n
06-29-2009, 06:01 PM
Ok my mistake. Interpolated shouldn't be ticked in Intuitions file. But lots of noise in the corners. Render time is ok at 5min per frame.
Anyone have an AC version of this scene that's flickerles nd good render time?

Intuition
06-29-2009, 06:03 PM
The interpolated FG may be set too low in my basic scene setup.

I am on vacation now in Wyoming. When I get back to LA I will try and setup a good Interp FG scene.

I can't say I have had any luck using this method with ACache.

toby
06-29-2009, 11:25 PM
here's my shot at AC, 7 min. 30 sec. on my machine, which should be less than a minute on an 8 core. If I'm not rendering anything overnight I'll do the whole sequence, but a lot can be judged from the still and the render time.

toby
06-30-2009, 02:06 AM
Looks like I accidentally swapped primary and secondary ray numbers :P I'm rendering it overnight with the correction.
Also I should mention a litle trick I use, bake at 50% multiplyer, render at 100%. Goes a little faster.

geo_n
06-30-2009, 09:30 AM
Looks like I accidentally swapped primary and secondary ray numbers :P I'm rendering it overnight with the correction.
Also I should mention a litle trick I use, bake at 50% multiplyer, render at 100%. Goes a little faster.

keep us posted. 50% multiplier is same trick in kray. sometimes blobs of light and shadow appear.

toby
06-30-2009, 01:51 PM
only 64 frames finished, 8 - 9 minutes each, 9hrs total time, about a minute per frame on an 8core.


keep us posted. 50% multiplier is same trick in kray. sometimes blobs of light and shadow appear.
Are you sure it's bake at 50, render at 100? I figured it was a fluke, the render is actually slower if you bake at 100 and render at 50.

toby
06-30-2009, 02:37 PM
Dang it, need to be more careful. Left the min. spacing at 0.5, slowing it down needlessly.

toby
06-30-2009, 03:13 PM
Yep that's me, head up me arse most of the day :screwy:

Ok, it's not much faster with corrected settings. What's important here is that it didn't get an epidemic of fg points and kill the render times, and so far it's faster than the motion blur method *unless* your scene needs lots of motion blur passes anyway.

geo_n
06-30-2009, 08:39 PM
only 64 frames finished, 8 - 9 minutes each, 9hrs total time, about a minute per frame on an 8core.


Are you sure it's bake at 50, render at 100? I figured it was a fluke, the render is actually slower if you bake at 100 and render at 50.

yes. calculate in half resolution then render in full size. In kray there's multiplier since before. In vray change camera resolution to half size and bake irridiance and lightcache then render at full size. But the is scene dependent. Some scenes its obvious and you see light and shadow blobs.
I'm glad its in lw 9.6 now. :thumbsup:

geo_n
06-30-2009, 08:57 PM
This is another test for comparison. I forgot to set the codec. Its animation codec but still playable in qt. Almost 0 gi flicker and render under 4 mins. Vray sp3.
I'll tweak the file for better and faster results.

toby
07-01-2009, 12:00 AM
yes. calculate in half resolution then render in full size. In kray there's multiplier since before. In vray change camera resolution to half size and bake irridiance and lightcache then render at full size. But the is scene dependent. Some scenes its obvious and you see light and shadow blobs.
I'm glad its in lw 9.6 now. :thumbsup:
The reason I think it's a fluke in lw is because rendering at 50% has always been faster, but with the cache it's slower; and I've yet to get any artifacts from it, and you'd think that baking at 100% would give you better results at the same speed when rendering. Just seems backwards to me. The bake doesn't take very long anyway so I wouldn't do it at 50% if it didn't turn out to be faster at rendertime.

toby
07-01-2009, 12:05 AM
This is another test for comparison. I forgot to set the codec. Its animation codec but still playable in qt. Almost 0 gi flicker and render under 4 mins. Vray sp3.
I'll tweak the file for better and faster results.
I like your first one better, compared to test00.mov this is a lot darker, and when the green ball is at the back wall you can see an unrealistic ambient shadow even where the light's hitting it (?) This is using ambient occlusion?

geo_n
07-01-2009, 10:04 AM
The reason I think it's a fluke in lw is because rendering at 50% has always been faster, but with the cache it's slower; and I've yet to get any artifacts from it, and you'd think that baking at 100% would give you better results at the same speed when rendering. Just seems backwards to me. The bake doesn't take very long anyway so I wouldn't do it at 50% if it didn't turn out to be faster at rendertime.

That's opposite to my experience in lw. Rendering at 50% with cache is always faster for me and rendering at 100% multiplier is slower but with better quality especially contact shadows. And less blobby looking shadows, too. So that's not what you have in lw9.6? Maybe its a mac thing.
If theres many blurry reflective material with high AA and AS it can be slow so I use 50% multiplier depends on scene.

test00 was done with irridiance and lightache. Fast results but slight flicker and not so accurate lighting. the next was irridiance and bruteforce which has almost no flicker. But it tends to darken image and needs different setting. There's no occlusion. I guess that's how accurate it would look. I think vray is good at faking lighting with good render time but either way it all looks real which tends to make most vray render look the same.

Exception
07-01-2009, 10:56 PM
Radiosity multiplier should not be used with Animated Cache (set other than 100%). It's a known issue. You have a high chance of slowdowns, flickering and other nasty things. However using the normal cache it's fine.

geo_n
07-02-2009, 10:40 PM
Radiosity multiplier should not be used with Animated Cache (set other than 100%). It's a known issue. You have a high chance of slowdowns, flickering and other nasty things. However using the normal cache it's fine.

Good to know. I use normal cache with multiplier only when there's no time to render. And I notice there's light and shadows blobs in 50% multiplier.
How about AC for the test scene to make it fast and flickerless the way you showed in your tutorial? tobys test might be unsuccessful since he's using 50% multiplier with AC.

Exception
07-03-2009, 02:04 AM
How about AC for the test scene to make it fast and flickerless the way you showed in your tutorial? tobys test might be unsuccessful since he's using 50% multiplier with AC.

Highly probable. I know my guide is long, but it does clearly say "Do not use the GI resolution multiplier with animated GI. It causes increasingly longer render times and possible flicker."
There are times you might be able to get away with it, but I know it breaks the sample placement code, so I would never risk it. You can possibly get away with actually changing the camera resolution, but I have never tried that and I wouldn't be surprised if that also won't work.

geo_n
07-03-2009, 04:30 AM
Highly probable. I know my guide is long, but it does clearly say "Do not use the GI resolution multiplier with animated GI. It causes increasingly longer render times and possible flicker."
There are times you might be able to get away with it, but I know it breaks the sample placement code, so I would never risk it. You can possibly get away with actually changing the camera resolution, but I have never tried that and I wouldn't be surprised if that also won't work.

care to give lw test file a try? I'm having hard time with AC. :D

toby
07-03-2009, 03:47 PM
Ok this is the weirdness I was talking about

One image here was baked at 400% and rendered at 100% gi multiplier, it took 16 minutes. The other was baked at 25% and rendered at 100, it took less than 4 minutes. There is a slight difference in quality, but side-by-side you wouldn't see it, you have to toggle between them.

I tried 400% thinking that if it baked more points into the cache, it wouldn't have to add as many during the render. But I guess it calculates every point that's in the cache, so more slows it down.

toby
07-03-2009, 04:01 PM
Here's the anim

There is a little bit of flicker, but I would never bake as low as 25% except for this test. I think 50% will be ok, I'll try to have that done tomorrow. But if it breaks something like Exception says, I guess it's a gamble and not reliable. But it's worth a shot :jam:

geo_n
07-03-2009, 08:23 PM
Here's the anim

There is a little bit of flicker, but I would never bake as low as 25% except for this test. I think 50% will be ok, I'll try to have that done tomorrow. But if it breaks something like Exception says, I guess it's a gamble and not reliable. But it's worth a shot :jam:

Can you post scene file for comparison so I can try render time on my machine? If anyone is interested to test vray version its here, max 2009 vray sp3. caching is less than 15min, render time less than 3min per frame. shadow and light is acceptable already. This is not optimized scene and I only did quick setup. I never thought any renderer would be faster than finalrender but I think this is fast gi from vraysp3.
Ok I try lw AC test this time.:D

toby
07-03-2009, 08:55 PM
Can you post scene file for comparison so I can try render time on my machine?
Absolootly!

toby
07-04-2009, 04:10 AM
Ok, baked at 50% multplier instead of 25, time went from 3 minutes/frame to 5, but the same slight flicker is still there, in fact it's baked into the cache, I can tell because increasing the rays to 256 over 128 does not help.
There's a pop between frame 6 and 7 on the underside of the ball, and it's still there with much higher settings.

So much for that method! Unless you think it won't be noticeable in your scene, or if you settle for a *minor* speed improvement by baking at 75%, maybe more.

Exception
07-04-2009, 05:00 PM
care to give lw test file a try? I'm having hard time with AC. :D

Which one?
And what should be tested?

I'm not testing the multiplier with AC. I know it doesn't work. And using FG is probably also not a good idea. I'd stick to MC for sure. FG is deprecated.

geo_n
07-04-2009, 07:40 PM
Which one?
And what should be tested?

I'm not testing the multiplier with AC. I know it doesn't work. And using FG is probably also not a good idea. I'd stick to MC for sure. FG is deprecated.

I attached the scene. This is tobys scene file.He did use 100% multiplier and monte AC. Renders fast but the results have lots of flicker with GI-Box-Area_tg_V06 and GI-bake50_V07 videos he posted.

Just to test it. I rendered it on qc6600 and baking was less than 5min and render per frame less 2 min.

geo_n
07-04-2009, 07:55 PM
From tobys scene file I up the setting.
300 rpe
mps 4-100

This is result. Render time is the same with tobys lower setting of rpe and mps.
Still this has too much flicker in green ball area and ceiling.
What's better is shadows have become small noise instead of shadow blobs due to higher settings. Better definition. This is noticeable in the yellow objects center part.

geo_n
07-04-2009, 08:06 PM
I'm rendering out a scene with the object gi setting set to local and some stuff changed. Its a pain to go through this if planning to use AC:D.

toby
07-04-2009, 08:19 PM
Min. spacing of 4 will definitely render a lot faster, but what you lose in definition is substantial. The only definition you have now is from the area light, which may be ok, but it's not something I'm prepared to give up -

And both of ours are flickering pretty bad at about frames 58 - 62

But I don't see how our shadows are different, can you point out a few specific frames where I can see it?

toby
07-04-2009, 08:23 PM
I attached the scene. This is tobys scene file.He did use 100% multiplier and monte AC.
But I baked it at 50%, that's the problem.

geo_n
07-04-2009, 08:59 PM
Min. spacing of 4 will definitely render a lot faster, but what you lose in definition is substantial. The only definition you have now is from the area light, which may be ok, but it's not something I'm prepared to give up -

And both of ours are flickering pretty bad at about frames 58 - 62

But I don't see how our shadows are different, can you point out a few specific frames where I can see it?

Yes its flickering pretty bad even with monte AC and I baked mine at 100. I think frames 15-25 I see your vid have blobs of shadows compared to my higher rpe producing noisy shadows. Lol
My next test which is rendering is local mps 1-200 on some objects and even higher rpe locally. Again this is a pain to do but the scene has few objects.

toby
07-04-2009, 09:14 PM
Excellent idea, you could leave min. spacing at 4 but turn to 1 for the scissors for example -

Have you been studying Exception's guide? If you really want a handle on AC, the best thing to do is learn all he has to say.

erikals
07-04-2009, 09:55 PM
Call me a naive fool, but It would be very straightforward to write each GO frame to a cache, which, then, at the end of the sequence this cache is analysed, to detect generalised true lighting shifts, and eliminate blotchy, one-frame-wonder flashes due to the interpolation.
Then this cache is post processed to smooth it all out and the post processed one is used in the actual render. this would work for deformation and everything.

Thats what I would do if I were in charge.. in fact, I'm sure if you baked the whole scene out to image maps, you could pretty much replicate this procedure in a compositing program.

But yet they faff about with sticking sampling points to geometry and such things.

regarding compositing,

yes, but no, i tested this a while back, it somewhat works when shadows/colors are blurred, but if they are not, and the scene uses e.g. hard shadows, it will not morph/fade properly. unfortunatly... :/

in other words, shadows on e.g. a floor won't come out right.

geo_n
07-04-2009, 09:56 PM
Excellent idea, you could leave min. spacing at 4 but turn to 1 for the scissors for example -

Have you been studying Exception's guide? If you really want a handle on AC, the best thing to do is learn all he has to say.

I played with AC as soon as it came out and also read the guide after it was made available. But I have yet to get good results for AC. Lol. I turn to kray or vray for gi stuff.
Now we have public lw scene and anyone can try to test with AC. But thinking about it this is fixed camera and already there's problem. I set up kray and vray scene in less than 30min. No secret setup needed there. It just works. kray scene acceptable even with interpolated gi and vrays new anim prepass/ lightcache combo is more than acceptable even with no tweaks.

Exception
07-05-2009, 12:45 PM
Now we have public lw scene and anyone can try to test with AC. But thinking about it this is fixed camera and already there's problem. I set up kray and vray scene in less than 30min. No secret setup needed there. It just works. kray scene acceptable even with interpolated gi and vrays new anim prepass/ lightcache combo is more than acceptable even with no tweaks.

LW's solution is definately not comparable to Kray and Vray's solutions. I see it more as a little extra feature that works in some cases. It doesn't work in many others and there's a long way to go until it does. But LW's engine is general-purpose and native, while Kray and Vray are specific purpose Gi engines that are third party. I'm happy we have access to Kray if we need to get stuff like this done efficiently.

I have trouble with this scene as well. I think the general way to get this flicker free with AC is: low grade MPs settings and very high RPE and SBR settings. The per-object tweaking is a good idea.

It might actually be worth while to send this into fogbugz and report it as being a specifically challenging scene for AC, good for testing. Sometimes problems like these are there not because of an impossibility in the code, but just because the specific circumstances had not been observed before to such an extreme.

toby
07-05-2009, 04:16 PM
Still this has too much flicker in green ball area and ceiling.
The underside of the ball is a serious problem area, seems like there's always a dark to bright pop at frames 4 to 5. I tested it with brute force and that shouldn't be there. I'm up to 256 ray primary and secondary (baked and rendering) to try and get rid of it, 12 minutes for a small limited region. I bet the blotches are due to the second bounce of the bright poly on the ceiling.

I tried the motion blur method again because it's so good at reducing blotches, but that didn't help, if fact it made me wonder if the second bounce is recalculated for each motion pass.

toby
07-05-2009, 09:41 PM
I can't get rid of it. Raising the angular tolerance to 45 helped, and so did lowering the max pixel space to 5, but even baking and rendering at 512 rpe, primary and secondary, it still flickers.

The second bounce *is* calculated with motion blur passes, I turned off the area light so the underside of the ball was lit by second bounce and nothing else. But with either method I'm having to use so many rays it's slower than brute force times, and the flicker's still there.

I give up! This scene def. needs to go to NT. I hope they have some improvements coming in 9.6.1, Core's gi is probably a year away or more.

allabulle
07-05-2009, 10:32 PM
I'm quite busy these days so I couldn't try the scenes with all of you, but I was paying close attention to what you were saying and testing. Thank you all.

Maybe for a scene like this we'd better get Kray. Is it ready for production yet? Does anyone have experience with Kray and is willing to share his thoughts? Or, better yet, anyone using Kray could be nice enough to try the scene tested here and post results with settings and machine specs and render times?

And now, I shut up for a while. :--)

Promise.

geo_n
07-05-2009, 11:31 PM
Exception - I tried low mps and high rpe and even higher rpe on some objects locally. It didn't get rid of flicker especially the ball behind it. The ceiling is ok now. Vray is not specific renderer imho. It is popular in archiviz but its as good as mray or finalrender or lw for general purpose cg. Plugins and max features work well thanks to good sdk. kray on the other hand has limitation similar to any third party render for lw.

toby - yeah I give up too. This is many attempts at AC since it came out and I think the trouble of setting it up just to get subpar results is a waste. I'll wait for Intution to post an optimized scene though its not AC it might be better to use his gi trick.

allabulle - I posted my first kray test here
http://www.newtek.com/forums/showpost.php?p=897543&postcount=77
Now I test higher setting but still render time is nearly the same.
It is interpolated and has limitations but I think that's acceptable more than what AC had to offer. Just use kray low/medium setting all round and set fullscreenAA to get motion blur. Didn't need to change anything from Intuitions file.

allabulle
07-06-2009, 12:02 AM
Ups, my bad. Thank you for posting it again and offering the link. I should pay more attention specially when I'm that tired. Thanks geo_n.

toby
07-06-2009, 12:28 AM
How fast was the Kray render? Looks just about perfect

Exception
07-06-2009, 01:38 AM
Exception - I tried low mps and high rpe and even higher rpe on some objects locally. It didn't get rid of flicker especially the ball behind it.

So far with this strategy I can't get rid of flciker either.
I tried an RPE of 4000 and a SBR of 500, and it still flickers a bit. I was trying a combination of 1000 RPE with a 4 pass motion blur trick which probably would have worked. That said, it's an ugly way to do it and very unsatisfactory.


Vray is not specific renderer imho. It is popular in archiviz but its as good as mray or finalrender or lw for general purpose cg. Plugins and max features work well thanks to good sdk. kray on the other hand has limitation similar to any third party render for lw.

I think Vray is highly specific. It's a dedicated purpose GI renderer. I don't see how more specific you can get. You can't do line rendering or off-the wall artistic non-realistic stuff in it like in LW's renderer. Although it's a little more flexible than Kray (not much), it's still a dedicated purpose renderer focused around a technology and an application. But that said, it's not helping us at this moment :)


I'll wait for Intution to post an optimized scene though its not AC it might be better to use his gi trick.

I think so too.


How fast was the Kray render? Looks just about perfect

Kray has some caveats as well though. Don't use specular shading and be careful with things like dissolves and morph maps.

Panikos
07-06-2009, 02:38 AM
All this testing is great. I am happy reading your conclusions even though we dont reach at the goal, which is consistent results. I wish Newtek considers all these seriously.
Pitty all these tests werent made during the beta phase.

Anyway, myself, I have spent very long time with GI since it was introduced in LW6.0 and later. Definitely LW grew up since that time and is moving towards the right direction, however slowly.

All LW features are portable, i.e you can render them on different machines and different platforms, on networks, etc GI is something very sensitive. Even interrupting it on the same machine and continue, "breaks" the consistency :(

geo_n
07-06-2009, 03:16 AM
Exception - what were your render times and what machine?
I think turn off gi in vray and use like scanline. Just like mray its fast in raytracing, etc, so instead of using max scanline for non photo render, people use mray instead or vray/finalrender if they have it. I post video made in vray, its like area lights look in lw.:D Not just archiviz for vray as many game/pachinco cg is done in max/vray.

toby - render time is around 2-3min per frame q6600. The flicker is very slight but I think with post its not noticeable anymore. One thing with kray I haven't seen non photo render. Doing only raytrace in kray the materials look very different to native render even when not using specular shading. Transparency setting is different. Maybe kray render is real.

toby
07-06-2009, 04:22 AM
Not just archiviz for vray as many game/pachinco cg is done in max/vray.
You won't believe this, but I had a real, full size pachinco machine when I was a kid (this type : http://www.asiandvdguide.com/toei/ty/05/ty5_fight_01.jpg )
Grandparents gave it to us, no idea where they got it. And I have no idea where it went, parents probably got rid of it :grumpy: oh well, I couldn't have dragged it around all these years -

Exception
07-06-2009, 06:46 AM
Pitty all these tests werent made during the beta phase.


I can asure you that they were, Panikos. Just because the feature is far from perfect now doesn't me we didn't find those problems way back then.
AC has a long way to go.

Panikos
07-06-2009, 11:57 AM
Yes, your web is a fantastic source

Larry_g1s
07-06-2009, 12:25 PM
Yes, your web is a fantastic sourceI agree. And if I can finish up what I need to do on my end for the LWCAD training, I can hand off the assets to Tom so he can do his Lighting/rendering video tutorials. :thumbsup:

Very soon guys.

erikals
07-07-2009, 06:20 PM
LW 9.6,
rendertime, 1 min per frame
note though, not motion blur, only AA

geo_n
07-07-2009, 06:49 PM
You won't believe this, but I had a real, full size pachinco machine when I was a kid (this type : http://www.asiandvdguide.com/toei/ty/05/ty5_fight_01.jpg )
Grandparents gave it to us, no idea where they got it. And I have no idea where it went, parents probably got rid of it :grumpy: oh well, I couldn't have dragged it around all these years -

You know I think they cost around 5-10000US? I'm not sure about antiques maybe more. :D
The machines now have lcd displays on the center where you watch and play pachinco at the same time.

geo_n
07-07-2009, 06:50 PM
LW 9.6,
rendertime, 1 min per frame
note though, not motion blur, only AA

Looks really good. :thumbsup: The best for lw so far. Normal cache? What machine spec?

erikals
07-07-2009, 06:52 PM
same render, but adding motion blur in post by using an after effects plugin.
(revisionFX motion blur)

erikals
07-07-2009, 07:09 PM
Looks really good. :thumbsup: The best for lw so far. Normal cache? What machine spec?

no animated cache here, only Final Gather + interpolated.
cpu used is a Quad 2.6GHz

i never thought it would work, but it is good to be wrong at times http://forums.cgsociety.org/images/smilies/smile.gif

i tested quite a bit, here is what seemed to work good,
you might want to try increasing the RPE and SBR, though i didn't see too much change, most of the flicker in the video is the video compression itself and the low AA setting.

i just hope it works this good for other scenes as well...

geo_n
07-07-2009, 08:10 PM
no animated cache here, only Final Gather + interpolated.
cpu used is a Quad 2.6GHz

i never thought it would work, but it is good to be wrong at times http://forums.cgsociety.org/images/smilies/smile.gif

i tested quite a bit, here is what seemed to work good,
you might want to try increasing the RPE and SBR, though i didn't see too much change, most of the flicker in the video is the video compression itself and the low AA setting.

i just hope it works this good for other scenes as well...

Yes, not using AC is good for lw. That's why I have not had use for AC really. i thought I'm not doing something right with AC but seems it really is not worth it.

toby
07-07-2009, 08:16 PM
Good Lord! FG is 3 times faster than MC?? Since when? I thought they were nearly the same, so I never bothered to try it! That many rays makes MC just as slow as brute force.

Well done sir!
Gotta see how this does with mo-blur though, too many problems with post-blur for me.

erikals
07-08-2009, 09:30 AM
Good Lord! FG is 3 times faster than MC?? Since when? I thought they were nearly the same, so I never bothered to try it! That many rays makes MC just as slow as brute force.

Well done sir!
Gotta see how this does with mo-blur though, too many problems with post-blur for me.

...doing a test now and it looks to work very very well with photoreal motion blur http://forums.cgsociety.org/images/smilies/smile.gif

geo_n
07-08-2009, 09:51 AM
...doing a test now and it looks to work very very well with photoreal motion blur http://forums.cgsociety.org/images/smilies/smile.gif

Something more to try is more than 1 light. lw radiosity tends to miss and hit solution when there's more lights. Flicker happens.

erikals
07-08-2009, 01:02 PM
Something more to try is more than 1 light. lw radiosity tends to miss and hit solution when there's more lights. Flicker happens.

two lights, no flicker.
most flicker here is due video compression.

photoreal motion blur,
rendertime per frame,... 2min 20sec

erikals
07-08-2009, 01:18 PM
...using the same settings, but switching to Monte Carlo, the rendertime per frame increased to,... 4min 10sec

i find that strange, as the two renders did look 99% identical

toby
07-08-2009, 02:56 PM
8 min. per frame on my old boat anchor = less than 1 min. on 8core
6 pass mo-blur, .07 as
96x70 rpe
50% multiplier
I think we're done here! Good to know what works.
I'll render the rest tomorrow

toby
07-08-2009, 04:12 PM
Ahhh... but how does it work with deformations?
I'm going to test that next but since it's not cached it should work, may have to turn on directional rays...

GraphXs
07-08-2009, 05:49 PM
This is great stuff! Glad to see a good solution with good times. What about glass? If FG is used doesn't it switch to MC to calculate the rays on the glass?:thumbsup:

toby
07-08-2009, 06:35 PM
Ahhh... but how does it work with deformations?
Seems to work fine

*but*

tyring to see what happens with glass - I never noticed before - that lw's gi doesn't get shadowed by anything that has any transparency. One image here is with .01 transparency, no other change. All types gi, directional rays on or off.

* nevermind, forgot the Use Transparency option

toby
07-08-2009, 07:05 PM
( what the H happened to bones? )

erikals
07-08-2009, 07:06 PM
deformations work good, renders fine...
however deformed objects with displaced polygons does increase rendertime alot.

it might seem strange, but the GI time doubled when i applied this deformation.

however, scenes without flat surfaces (iow, no architectural scenes), can have more flickering,
as it will not be all that visible.

this means renders of e.g. nature scenes can have lower GI/AA/MB settings.

erikals
07-08-2009, 07:20 PM
( what the H happened to bones? )

heh, new feature? http://forums.cgsociety.org/images/smilies/smile.gif

toby
07-08-2009, 07:31 PM
deformations work good, renders fine...
however deformed objects with displaced polygons does increase rendertime alot.

it might seem strange, but the GI time doubled when i applied this deformation.

however, scenes without flat surfaces (iow, no architectural scenes), can have more flickering,
as it will not be all that visible.

this means renders of e.g. nature scenes can have lower GI/AA/MB settings.
Yea I think it's all good.

Intuition
07-08-2009, 08:04 PM
Wow, nice one. Didn't even think to use the %50 trick like in Kray.

I am back in town now. :) But it looks like you guys have a nice solution. :D

Intuition
07-08-2009, 08:05 PM
( what the H happened to bones? )

Oh, I saw that last Oct working at Universal (BSG VFX), I think you can change the display size for the bones so that they don't look fat.

Intuition
07-08-2009, 08:07 PM
two lights, no flicker.
most flicker here is due video compression.

photoreal motion blur,
rendertime per frame,... 2min 20sec

So wait, PRmotion blur will blend the samples now?

toby
07-08-2009, 09:31 PM
So wait, PRmotion blur will blend the samples now?
I don't think so, it's just not always noticeable. Or he's rendering 1 thread :)

toby
07-08-2009, 09:35 PM
Wow, nice one. Didn't even think to use the %50 trick like in Kray.

I am back in town now. :) But it looks like you guys have a nice solution. :D
Yes, our work is done here :D

The real revelation (erikals') was to use FG instead of MC - makes a huge difference in this scene, I'm still wondering why. Maybe because everything is completely enclosed in geometry, and it's rendering through one of the walls?

Intuition
07-11-2009, 02:27 PM
Yeah, I think I mentioned that you can use this method with FG, and you can even use FG interpolated as well. The %50 trick would make tuning even quicker for sure though. :D

toby
07-11-2009, 05:13 PM
Is there a "final" scene posted here that we can check?

All I can say, is THANK YOU for the great testing you guys did here.

It IS MUCH appreciated!!
Oh, whoops :) here's mine

however, I tried these settings in another scene and it wouldn't stop flickering until I used the anim cache. So as always, it 'depends on the scene' :bangwall:;D

erikals
07-12-2009, 10:42 PM
...however, I tried these settings in another scene and it wouldn't stop flickering until I used the anim cache. So as always, it 'depends on the scene'

toby, wondering, could you post a screengrab of the scene, or a similar scene.

Hieron
07-21-2009, 04:59 PM
Oh, whoops :) here's mine

however, I tried these settings in another scene and it wouldn't stop flickering until I used the anim cache. So as always, it 'depends on the scene' :bangwall:;D

I'm probably stating the obvious here, but it baffled me for a while. Using say a dome light with low Q setting and letting the AS figure it out is great, but messes up GI samples for me whatever I do. I guess AC would help there.

Just a thought...

probiner
07-21-2009, 07:01 PM
Nice info in this thread.
Thanks

Carm3D
08-04-2009, 03:05 PM
I love this "spinning radiosity" trick.. I wish I had seen this thread months ago. I'm gonna lower my light quality settings and give them a spin too... Might as well. :)

erikals
08-04-2009, 10:01 PM
was testing a bit here too,... for interior rendertypes, which tend to take much longer to render to reduce flickering...

http://www.newtek.com/forums/showthread.php?p=911643#post911643