PDA

View Full Version : New blur in action



grangerfx
08-01-2006, 11:44 AM
Below is an example of the new 3D motion blur in action. This is the Brunswick Pins classic content scene with the camera mode set to Perspective and the blur mode set to the new 3D blur.

-Mark Granger

Elmar Moelzer
08-01-2006, 11:52 AM
Looking very, very good Mark!
:)
Can you explain a bit more about how it works, especially compared to LWs classic MotionBlur?
CU
Elmar

grangerfx
08-01-2006, 12:02 PM
Each sub frame of the classic motion blur is rendered at the same sub-frame time. In the new motion blur, each pixel is rendered at a different sub-frame time. The more samples or sub frames the smoother the blur appears. The new blur can also do things like curved blur streaks due to motion paths and/or rotation of objects. Setting up the new blur is as simple as selecting the new blur mode in the camera panel and rendering.

You can also select the streak length (same as the shutter angle). The default is 50% representing half a frame of time. You can enter any value however. If you make this number very large, you can actually capture your entire animation in a single frame.

-Mark Granger

Shulmanator
08-01-2006, 12:07 PM
I know you guys probably aren't finished yet, but with what you have so far, can you do a side-by-side comparison of the same frame, between the new 3D blur, and the classic stepped blur cranked up to the closest similarity to the 3D blur, and state the render times for each?

Thanks.

ufo3d
08-01-2006, 12:09 PM
hi Mark, any news about rendering with GI?

JensD
08-01-2006, 12:14 PM
Hi,

That is good news! Can you give us an idea of the time it takes to render a frame compared to the old motion blur?

Greetings

Jens

digital verve
08-01-2006, 12:31 PM
Nice. That is a big improvement.

Earl
08-01-2006, 12:46 PM
Thanks for posting the example Mark! From what I can see so far, it seems like a great improvement. I look forward to seeing more examples, and testing it out for myself. :lightwave

Elmar Moelzer
08-01-2006, 12:53 PM
Hey Mark!
This truely sounds awesome!
So this means that we can use the spinning light trick with the new motionblur? Or do we have to combine the new and the old multipass blur for that (which would be fine with me too).
In any case the results look just great!
Cant wait to get my hands on this!
CU
Elmar
PS: Will it work with Volumetrics?

Shulmanator
08-01-2006, 12:55 PM
So this means that we can use the spinning light trick with the new motionblur? Or do we have to combine the new and the old multipass blur for that

Good question. I wanna know as well.

grangerfx
08-01-2006, 12:55 PM
The current implementation of the new motion blur renders slower than the classic motion blur. There are several opportunities for optimization of the new motion blur algorithm so I expect the rendering times to decrease considerably before release.

Attached is a rendering of the the Brunswick Pins scene using the classic motion blur and with the same level of anti-aliasing as the first image. You can see the stepped blur in the textures on the rapidly moving pins.

-Mark Granger

jayroth
08-01-2006, 12:56 PM
I know you guys probably aren't finished yet, but with what you have so far, can you do a side-by-side comparison of the same frame, between the new 3D blur, and the classic stepped blur cranked up to the closest similarity to the 3D blur, and state the render times for each?

Thanks.

Its a little too early for timing comparisons just yet, as no optimization work has been done. I can say that the times are similar now, but that we expect the new blur routines to be faster overall...

Jay Roth

JamesCurtis
08-01-2006, 12:59 PM
Any idea of when it'll be available? Not in LW9 as of yet, right?

grangerfx
08-01-2006, 01:00 PM
So this means that we can use the spinning light trick with the new motionblur? Or do we have to combine the new and the old multipass blur for that (which would be fine with me too).


We will have to implement the new blur feature by feature. Light sources should happen soonest so spinning lights should work OK when we ship this feature. I think that volumetric lights will work too but other volumetrics may not. We will need to expand the plug-in APIs to allow third party volumetrics like yours to work correctly (if you can). At worst case you can just combine the old and new blur as you suggest. That is working currently. You still get the benefit of the new blur for camera motion for free though!

-Mark Granger

KillMe
08-01-2006, 01:01 PM
hmmmmm flicking between the to images - the difference between the two is marginal to say the least

Celshader
08-01-2006, 01:01 PM
Attached is a rendering of the the Brunswick Pins scene using the classic motion blur and with the same level of anti-aliasing as the first image. You can see the stepped blur in the textures on the rapidly moving pins.

I'm not sure, but I think you attached the same "3D blur" image by mistake... :o

Elmar Moelzer
08-01-2006, 01:01 PM
Hey Mark!
I am fine with it, as long as it works and does not produce artefacts...
We will certainly look into making VoluMedic work with it as soon as you get us the SDK features for that.
CU
Elmar

grangerfx
08-01-2006, 01:03 PM
And in case you are wondering, yes I started implementing this feature after we shipped LW 9. That is why we have not had time to start getting into optimization and meaningfull rendering time comparisions yet.

grangerfx
08-01-2006, 01:08 PM
(deleted the original post) See the next post for a better comparison between the two blur algorithms.

grangerfx
08-01-2006, 01:15 PM
I think that the anti-aliasing level was too high to really see the difference. Here is a pair of images to compare the classic and new blur algorithms with medium anti-aliasing. The noise level is higher but you can clearly see the differences in how the two algorithms work.

-Mark Granger

Shulmanator
08-01-2006, 01:17 PM
spinning lights should work OK when we ship this feature.
-Mark Granger

If this is true, then this is something that Lightwave will be able to do, render-wise, that I don't think can be done with Maya's Mental Ray.

Shulmanator
08-01-2006, 01:19 PM
Here is a pair of images to compare the classic and new blur algorithms with medium anti-aliasing. The noise level is higher but you can clearly see the differences in how the two algorithms work.

-Mark Granger


Well, I suppose grainy blur is less distracting to the eye than steppy blur. So that's certainly an improvement.

Weepul
08-01-2006, 01:24 PM
Ah, so it's stochastic now instead of stepped. A decent improvement, and something I've been hoping will come to LW. It's still ray-traced (or "brute force") blur, though, so a more smoothly-working image filter would be very appreciated.

I'd actually set up a node tree to create a surface that renders stochastic DOF blurring already in 9. :D

Something I'm wondering about: if each pixel has its own unique position in time (or a selection from a greater number of points in time than with the classic motion blur), how will that be reconciled with scenes where deforming, subdividing, or moving the geometry takes quite a bit of time to calculate?


I'm not sure, but I think you attached the same "3D blur" image by mistake... :o
Ouch! 8/ They're similar, but not identical.

grangerfx
08-01-2006, 01:34 PM
Something I'm wondering about: if each pixel has its own unique position in time (or a selection from a greater number of points in time than with the classic motion blur), how will that be reconciled with scenes where deforming, subdividing, or moving the geometry takes quite a bit of time to calculate?

Suffice it to say I have ways to solve those problems.

-Mark Granger

Bytehawk
08-01-2006, 01:43 PM
That is truly an amazing progress. Congrats on this.

would love to get my mits on this.

jayroth
08-01-2006, 01:44 PM
Well, I suppose grainy blur is less distracting to the eye than steppy blur. So that's certainly an improvement.

Note that the quality of the image was dialed down to show the you the differences at corresponding levels to the original blur algorithm. Unlike the old blur, which is directly linked to the anti-aliasing engine, the new blur is independent of it. Therefore you will have separate sampling settings for AA, motion blur, and DOF once we are done. You will be able to control the noise quality.

As has been mentioned previously, its way too early to consider timing comparisons, other than to note that in this early stage, the results are similar, with the new blur taking a tad longer when the quality is dialed up. Once we optimize, it will be quite a different story.

Jay Roth

connerh
08-01-2006, 01:57 PM
Hmm... I'll bet writing a little lscript to blur the... blur... in post wouldn't be too difficult to write... heck, I'm sure even adding vector blur over that with say... 5%-10% blur would knock that grain right out. Assuming, of course, that the vector information is made available to plugins again.

But on another note, great work guys! This has been something that I've been craving for quite a while now.

monovich
08-01-2006, 02:00 PM
looks pretty good. looking forward to testing it out when you guys unleash it to the mongrol hoards in the Open Beta.

KillMe
08-01-2006, 02:32 PM
any chance we could make it so motion blur can be applied on a per object basis - so you can have alot of blur on one object - not much on something else

take making a rain effect - need wads of motion blur on to make it look right but then everything else has 3000% motion blur too which is jsut not workable so have to render in passes - and while i know i probally should be doing that i'd rather not 90% of the time

while we on that subject how about a per object blur - i have this technique for making nice cartoony clouds but requires again wads of blur ( not of the motion variety this time ) using precision blur image filter or something like that anyway again measn rendering in passes when it shouldn't be needed

Phil
08-01-2006, 02:35 PM
any chance we could make it so motion blur can be applied on a per object basis - so you can have alot of blur on one object - not much on something else

take making a rain effect - need wads of motion blur on to make it look right but then everything else has 3000% motion blur too which is jsut not workable so have to render in passes - and while i know i probally should be doing that i'd rather not 90% of the time

I think you might be able to do this with Relativity 2 - you can manipulate the motion to give that kind of effect by playing with the time parameter. There was content showing this, for original customers of Prem, but I'm still not sure if this can be released in any way. I've emailed Prem, but had no response; there is no word from NewTek either. I'm not sure who it belongs to after the acquisition....

Current 9.0 'bonus' content is lacking substance in this area.

JensD
08-01-2006, 05:00 PM
As i understand it the classic motion blur is connected to the system off renderpasses like the anti aliasing system in Lightwave.

With the new motion blur and depth of field, is there a chance off getting better, modern AA routines which do not rely on renderpasses.

Sorry for my bad english.

Greetings

Jens

Shulmanator
08-01-2006, 05:39 PM
I think you might be able to do (motion blur per object) with Relativity 2 - ...

In Maya, turning an object's motion blur on and off is as simple as a checkbox. I think Newtek should look into this.

Dave Jerrard
08-01-2006, 05:39 PM
So this means that we can use the spinning light trick with the new motionblur? Since the spinning light relies on motion blur to do its 'magic', this can only make it look better. Right now, the trick really works best with 9 or more passes, which is enough to hid the multiple shadow effect you can get with lower AA levels. With this new method, you shouldn't see any individual shadows anymore. And because of that, part of the equation for the spinning light coudl actually be removed because it's not needed anymore.

Right now, the formula for getting the best blur is: (720/MB)-(720/MB/P), where MB is the blur amount expressed as a decimal (50%=0.5) and P is the number of render passes (double it if you're using Dithered Blur). The second part is used to offset things a bit so the first and last render pass will not be identical, which they would be with (720/MB).

Since this new blur doesn't rely on full frame time slices anymore, you won't run into a situation where two full passes are identical. Any pixel would theoretically have an equal probablity of recording the beginning of the frame, or the end, or any point in time between. As Jay mentioned earlier, the new blur works with reflections and shadows, so that tells me that a spinning shadow should look a lot better as well, so in effect, we will have an improved (and easier to set up) spinning light trick. :thumbsup:


He Who Still Remembers Looking At A Wooden Gate When He Thought This Sucker Up.

Dave Jerrard
08-01-2006, 06:17 PM
I think that the anti-aliasing level was too high to really see the difference. Here is a pair of images to compare the classic and new blur algorithms with medium anti-aliasing. The noise level is higher but you can clearly see the differences in how the two algorithms work.

-Mark Granger
I see a slight shift in positions of the objects. Does the new blur work by blurring both forward and backward in time, or is it still blurring backward like the LW has since version 3?


He Who Picks Up On Some Of The Weirdest Stuff.

jayroth
08-01-2006, 06:22 PM
Actually, the new blur is currently rendering forward in time.

Jay Roth

RedBull
08-01-2006, 06:51 PM
Nice it's looking good to me.... Many people have asked for better MB in particular for many years. I agree per object MB would be cool.

It definately showing some inhouse work and development, and it's even better to have Mark and Jay, and testers like Dave all discussing it's implentation and story on public forums. This only creates great energy for new and old users alike. Well done, and keep it up.

UnCommonGrafx
08-01-2006, 10:19 PM
Jay,
So, when will these things begin to trickle out? Will they be available to open-beta participants prior to being rolled out in the next 9.x release?
How's the wife and expected one? Hope all is good and may it become even better.

Sensei
08-01-2006, 10:48 PM
I think that the anti-aliasing level was too high to really see the difference. Here is a pair of images to compare the classic and new blur algorithms with medium anti-aliasing. The noise level is higher but you can clearly see the differences in how the two algorithms work.

Can you render scene with solid color on the table, instead of wood? I thought that Jay said in the note that this 3d motion blur will be visible in reflections.. but on attached renders, I don't see it..

grangerfx
08-02-2006, 01:29 AM
With the new motion blur and depth of field, is there a chance off getting better, modern AA routines which do not rely on renderpasses.
Yes! As a matter of fact I could not use the original multi-pass anti-aliasing method with the new blur so I had to go ahead and implement a fully random sampled anti-aliasing along with the new motion blur. This should eliminate artifacts like those that occur with thin parallel lines.

-Mark Granger

grangerfx
08-02-2006, 01:46 AM
Can you render scene with solid color on the table, instead of wood? I thought that Jay said in the note that this 3d motion blur will be visible in reflections.. but on attached renders, I don't see it..
Will this do?

-Mark Granger

JensD
08-02-2006, 02:01 AM
Yes! As a matter of fact I could not use the original multi-pass anti-aliasing method with the new blur so I had to go ahead and implement a fully random sampled anti-aliasing along with the new motion blur. This should eliminate artifacts like those that occur with thin parallel lines.

-Mark Granger

That is great news.

Jens

gerardstrada
08-02-2006, 03:24 AM
There are many other effects that take advantage of the Dave Jerrard's spinning light trick to be achieved in quicker times, effects like ambient occlusion, radiance effect, volumetric effects (like clouds, lights, liquids, smoke, fire), chromatic aberrations in reflections and refractions, hand painted "look" (pencil, characol, watercolors, etc), soft reflections, soft refractions, SSS, Dof, etc... for what would be excellent if this new Motion Blur could affect the same features as the classic Motion Blur does. In cases in that this is not possible, is very convenient to be able to combine the old and new blur, I think. This news looks pretty cool :) Thanks Mark!



Gerardo

gerardstrada
08-02-2006, 03:57 AM
In Maya, turning an object's motion blur on and off is as simple as a checkbox. I think Newtek should look into this.


Just turn ON or OFF object's motion blur may be a little limited. But we can adjust the percentage of motion blur per object with Object_Blur (from Worley) and per any item with TB_BlurControl; I remember also Independent_Motion_Blur (by Morgwn McCarty from LW 5x days), this plug still works in LW 8.x and it also works for any item. If I remember well, TB_BlurControl and IMB are free.



Gerardo

Darth Mole
08-02-2006, 04:16 AM
I kow this is a bit sacreligious (and slightly Off-topic), but have you guys ever used Realsmart Motion Blur? It's a plug-in (I use it in After Effects). Darn thing's brilliant: I just render with low AA to get rid of jaggy edges and then run it through AE. Saves loads of time.

The image below is the same scene rendered with five lots of full-screen AA and LW motion blurring, the other has just two PLD passes (with adaptive sampling on) and then had RSMB applied.

The second one looks better and took way less time!

Dave Jerrard
08-02-2006, 05:09 AM
I kow this is a bit sacreligious (and slightly Off-topic), but have you guys ever used Realsmart Motion Blur? It's a plug-in (I use it in After Effects). Darn thing's brilliant: I just render with low AA to get rid of jaggy edges and then run it through AE. Saves loads of time.Nice.

But the big question... How well does it handle transparency, reflections and shadows? Every 2D post process I've ever seen has failed on these. At least without a ton of tweaking and compositing trickery.


He Who Was Pleasantly Surprised To See XDof Doing Really Good DOF Blurs.

Darth Mole
08-02-2006, 06:54 AM
Ah... good question. I dunno, really. I guess I'll have a play. Since it's based on the movement of pixels, it should work fine - if it moves, it'll blur it. Could you suggest a scene to try it on?

He Who Has Just Stolen DJ's Catch Phrase.

:-)

kattkieru
08-02-2006, 08:43 AM
any chance we could make it so motion blur can be applied on a per object basis - so you can have alot of blur on one object - not much on something else

Not to bring up other software, but Blender solved this in a very interesting way by allowing Blur to be done in composite nodes. You can adjust / fake vectoral blur using a node which generates the amount of distance the pixels have traveled. The fun part about this is that you can animate it so that characters or objects blur/flicker on certain frames, like when Superman starts into his super speed.

I'm not (yet) a proud owner of LW9, but when I become one per-object blur from within Lightwave, without external plugins free or otherwise, is something I know I'll enjoy having. Having the amount be animateable would also be really cool.

Elmar Moelzer
08-02-2006, 09:02 AM
Postprocesses dont work well with transparencies and overlapping elements.
You would have to render them out separately and then composit them too.
I know that some people especially at larger studios are doing that, but IMHO a real 3d- motionblur or real 3d- DOF will always look better (if the renderer does it well which LWs renderer will do). Not to mention that it is easier to handle for smaller studios, or single people.
Let me give you an example where 3d- motionblur will look better than 2d (and an example that will benefit greatly from the new improved motionblur):
Lets say the Mustang plane from LWs classic content is flying towards the screen at a very fast speed.
Now a few things are happening here:
1. The plane is moving towards the camera and there is very little motion to the side.
This is already something where many post processes will have problems.
2. The perspective distortion will change as the plane comes closer.
This means that the nose will "cover" more of the plane the closer the plane comes to the camera. This will be very hard to simulate with a 2d- blur and I dont think that any of the existing 2d- motionblurs can get that right, unless they are comparing neighbouring frames and even then it will be difficult if the motion is fast.
3. The propeller, if it is not rendered separately, will cover part of the plane. It is impossible for a post process to recreate the covered part of the plane.
4. Again the propeller. In a postprocess you only have information about the current vector or a pixel. A vector is a straight line not a curve. So the pixels of the propeller will be blurred in a straight line, which will look wrong (ever tried LWs vectorblur on a propeller?).

Some of these negative effects of a post motionblur might be solvable with more renderlayers and a postprocess that also takes neighbouring frames into account, but not all of them.

Now for DOF, you have even more problems.
1. Coverage
If you dont render out objects that are partially covered by others in separate layers, the post process will not be able to recreate what was behind the foreground object. This will always lead to artefacts.
Now if you have a complex object that is very deep and has foreground parts that partially cover parts of the object further back it will be quite a task to separate these parts out into separate layers for a post DOF.
2. The Zbuffer: a zbuffer ignores transparencies. This means that a semi transparent ball will appear solid in the zbuffer. This also means that objects behind it will not be blurred correctly (will be blurred with the zvalue of the transparent object in front of them). This is a problem that XDOF has as well.
The solution would be to have transparent pixels in a separate buffer. But this again would not help if you had multiple semi transparent items partially covering each other (e.g. a transparent ball, behind a transparent ball).
And so on.

IMHO the improvements that NT makes to the DOF and Motioblur are awesome and will help LWs renderer get back on top of the pack of rendering solutions out there.

MR e.g. does not do curved 3d- motionblur on shadows from what I know. So spinning lights wont work with it. If NT gets that right in LWs renderer (or at least allows to combine the classic with the new motionblur), LWs renderer will have a huge advantage here.
I am really looking foreward to these new features in LW!
:I_Love_Ne
CU
Elmar

Matt
08-02-2006, 09:20 AM
Will this do?

-Mark Granger

Certainly will! Looks like an awesome improvement.

BTW: Loving the upfront previews, very welcome believe me! Thanks Mark (and Jay!)

:thumbsup:

Earl
08-02-2006, 09:37 AM
That ant is a really great example of reflected motion blur. Thanks Mark!!

Lightwolf
08-02-2006, 11:18 AM
Do I get this right, LightWave will finally be able to render with Mental Ray-ish AA? By this I mean no more render passes and the whole AA'd frame rendered in one pass, motion blur or not.
I doubt it. It seems that there will still be passes, but the samples for every pixel will be dithered in time.

Mind you, internally mr does the same thing to every tile, you just don't see it since it displays the tile after the passes/samples have been accumulated and filtered.

Cheers,
Mike

Shulmanator
08-02-2006, 12:00 PM
Just turn ON or OFF object's motion blur may be a little limited. But we can adjust the percentage of motion blur per object with Object_Blur (from Worley) and per any item with TB_BlurControl; I remember also Independent_Motion_Blur (by Morgwn McCarty from LW 5x days), this plug still works in LW 8.x and it also works for any item. If I remember well, TB_BlurControl and IMB are free.



Gerardo


I've never heard of these plug ins. Has anyone else used them and liked the results?

grangerfx
08-02-2006, 01:39 PM
Do I get this right, LightWave will finally be able to render with Mental Ray-ish AA? By this I mean no more render passes and the whole AA'd frame rendered in one pass, motion blur or not. For regular everyday use, no strings attached?
You will be able to select the number of passes and number of samples per pass independantly. For most scenes, a single pass with enough samples for anti-aliasing and motion blur will work fine.

I am leaving the option to render multiple passes in because there will be some things that will not work with the new motion blur. We will be implementing more of these over time but there will always be some feature, such as a legacy plug-in, that will not be able to fully support the new motion blur. In these cases, the multi-pass method is best especially when combined with the new blur for all the other features in the scene.

-Mark Granger

EyelandArts.com
08-02-2006, 02:17 PM
I've never heard of these plug ins. Has anyone else used them and liked the results?

Yeah, I have used TB's in the past and they work great and transparently. I seem to remember you could dial in the amount. This has been around forever. its a shame some people still dont know about it.

Emmanuel
08-02-2006, 02:27 PM
Thanks Mark, really great things You show us here !

Elmar Moelzer
08-02-2006, 02:44 PM
Hey Mark!
Sounds really awesome to me!
Love the way you have implemented this. Should also help with backwards and plugin- compatibility!
CU
Elmar

Ztreem
08-03-2006, 05:19 AM
This is great news! I like the way it is presented in an early stage here at the forum for everyone to join and see the progress, thanks!

archiea
08-05-2006, 06:40 PM
Thanks Mark...

Do you have a sample showing the blur curve along a path?

papou
08-06-2006, 09:20 PM
very great news.
hope it can be possible to optimise code to get speed increase when it can be done...

i would like to see spinning and aaliased grid exemple too, if its possible...

happy

adk
08-15-2006, 07:31 PM
... this is great news & I'm sure I'm not the only one eagerly waiting to get my mits on this :thumbsup: :thumbsup: :thumbsup:

pixym
09-02-2006, 07:03 AM
I hope after these Blur and DOF major improvements, Archi Viz LW users will benefit from a very fast GI improvements for interior animation (and also exterior btw).
Thanks in advance Marc and Jay ;)

Edit: BTW also improve area light rendering speed.

kmaas
09-02-2006, 12:01 PM
OK, not quite on the subject, but is there a way to save out AA passes so you can render at a higher AA setting later? That's something I've pretty much always wanted - save out a 1-pass frame that looks awful, then, if you have time, be able to hit F9 and have it use the saved first pass. Can it do that? You could get so much better results on a rendering time budget if it could.

bryphi7
09-02-2006, 12:06 PM
OK, not quite on the subject, but is there a way to save out AA passes so you can render at a higher AA setting later? That's something I've pretty much always wanted - save out a 1-pass frame that looks awful, then, if you have time, be able to hit F9 and have it use the saved first pass. Can it do that? You could get so much better results on a rendering time budget if it could.

Yah...its called buy Fprime:) does exactly what you are asking...

T-Light
09-02-2006, 12:07 PM
That's an fprime only function I'm afraid.

ericsmith
09-02-2006, 12:30 PM
It's tricky, but if you can get your camera to shift by a half-pixel, then you could do this:

step 1. render a single no AA pass

step 2. shift the camera 1/2 pixel to the left and render. repeat to the right, and up and down. You've now rendered 4 passes, plus the one in step 1.

step 3. layer all 5 renders with each at 20% opacity. Now you have a single image with 5 pass AA

step 4. If you want 9 pass AA, render an additional 4 with by moving the camera up and left 1/2 pixel, then up and right, down and right, and down and left. Layer all of the images at 11.111% opacity.

If you're really clever, you could probably even pull off motion blur by rendering each pass at a subframe amount appropriate for the distance you want to blur.

This is all theoretical, but it would be a cool experiment. The tricky part is figuring out how to translate the distance of 1/2 pixel into a distance you could move the camera, and then rigging the camera up so that you can always move it paralell to the image plane. It's possible that the distance could be determined through some formula based on the pixel size of the render and the zoom amount.

Eric

T-Light
09-02-2006, 07:49 PM
ericsmith -

step 1. render a single no AA pass

step 2. shift the camera 1/2 pixel to the left and render. repeat to the right, and up and down. You've now rendered 4 passes, plus the one in step 1.

step 3. layer all 5 renders with each at 20% opacity. Now you have a single image with 5 pass AA

step 4. If you want 9 pass AA, render an additional 4 with by moving the camera up and left 1/2 pixel, then up and right, down and right, and down and left. Layer all of the images at 11.111% opacity.
Alternatively you could just antialias with the WaveFilterImage plugin :D

Captain Obvious
09-02-2006, 08:08 PM
It's tricky, but if you can get your camera to shift by a half-pixel, then you could do this:

step 1. render a single no AA pass

step 2. shift the camera 1/2 pixel to the left and render. repeat to the right, and up and down. You've now rendered 4 passes, plus the one in step 1.

step 3. layer all 5 renders with each at 20% opacity. Now you have a single image with 5 pass AA

step 4. If you want 9 pass AA, render an additional 4 with by moving the camera up and left 1/2 pixel, then up and right, down and right, and down and left. Layer all of the images at 11.111% opacity.

If you're really clever, you could probably even pull off motion blur by rendering each pass at a subframe amount appropriate for the distance you want to blur.

This is all theoretical, but it would be a cool experiment. The tricky part is figuring out how to translate the distance of 1/2 pixel into a distance you could move the camera, and then rigging the camera up so that you can always move it paralell to the image plane. It's possible that the distance could be determined through some formula based on the pixel size of the render and the zoom amount.

Eric
I actually experimented with that a long time ago. I'm too tired right now to search to find the thread, though... I got it working pretty good, actually. I rendered fully raytraced depth-of-field in Lightwave 8.2, something not even in the feature set. Cool, huh? :)

My original idea was to use it for more efficient distributed rendering of single frames. The whole slice-it-up-into-segments is not a very good method, in my opinion. If you just animate the camera a bit and use Lightwave's distributed rendering, you should get very efficient distributed rendering of single frames.

kmaas
09-04-2006, 08:00 PM
Yah...its called buy Fprime:) does exactly what you are asking...

Yeah, yeah. Everyone seems to think it's the best thing since sliced bread, and it really would be nice (looks like an awful handy program), but I don't animate enough to merit buying it.

kmaas
09-04-2006, 08:01 PM
It's tricky, but if you can get your camera to shift by a half-pixel, then you could do this:

step 1. render a single no AA pass

step 2. shift the camera 1/2 pixel to the left and render. repeat to the right, and up and down. You've now rendered 4 passes, plus the one in step 1.

step 3. layer all 5 renders with each at 20% opacity. Now you have a single image with 5 pass AA

step 4. If you want 9 pass AA, render an additional 4 with by moving the camera up and left 1/2 pixel, then up and right, down and right, and down and left. Layer all of the images at 11.111% opacity.

If you're really clever, you could probably even pull off motion blur by rendering each pass at a subframe amount appropriate for the distance you want to blur.

This is all theoretical, but it would be a cool experiment. The tricky part is figuring out how to translate the distance of 1/2 pixel into a distance you could move the camera, and then rigging the camera up so that you can always move it paralell to the image plane. It's possible that the distance could be determined through some formula based on the pixel size of the render and the zoom amount.

Eric


Cool idea! I may just have to try that sometime.