What comes first ?

Motion Blur vs. DOF = What should you treat first ?

I was wondering what you think about the priority of these two features.
Do you usually add motion blur first and after you apply the DOF effect ?

Or do you do it the other way arround ?
Curious if you could do it both ways or if you run into any issues !

Thanks for any little help, hints, tipps and tricks and your experience managing these options !

cheers

Motion blur then depth of field would be my choice.

Or ask Chris Bond to re-release MotionDepthBlur from Awake.

i would do it the way that looks best, just that simple.
because there is no technical correct way to do it (besides the plugin that chad mentioned) you have to look what is the best result.

Or ask Chris Bond to re-release MotionDepthBlur from Awake.

I've been looking for something like that for a while now!
I tend to not use Motion vector kicked out from 3D anymore, because of issues between motion blur and depth.
Now I tend to apply DOF and then use some sort of pixel based motion blurin Fusion , like RSMB or Optical flow in Fu7. That seems to work most of the time.

Hmm.. so you mean, forget about the Vector Motion Blur ?
Better to use DOF and after that, just simple apply a non vector based pixel Motion Blur ?

More like "choose whatever looks least bad for your situation".

Ideally, you'd render out with motion blur and depth of field.

It really is whatever looks best for your situation type deal.
yes ideally you render out everything with motion blur and dof, but sometimes that gets you cornered into a spot where your only solution is to rerender the 3d, and you've lost the flexibility and speed of compositing.
I've just found that in some of my later projects i get results that are about as good with non vector based pixel motion blur. Now that Fusion Studio ships with optical flow,why not? ( though the new RSMB is said to be GPU accelerated which would be nice).

Speaking of that Awake plugin, have you seen Psyop's crypto matte? Looks like an ID matte on steroids, with AA, motion blur, transparency, and Dof, and it's automatically generated. I'm guessing that with something like that you could still render everything in camera and still have quite a bit of flexibility in post.
http://www.psyop.tv/psyops-cryptomatte- ... 15-poster/

RSMB is GPU accelerated, but it uses it's own OpenCL manager, not Fusion's, which is OK, but misses out on some interoperability niceness. It's still going to have the same issues though, missing occlusions, etc., in addition to the optical flow uncertainty.

That cryptomatte is so clever, it's amazing that it isn't standard. Requires a lot of setup from the renderer, it's not just a shader, and it requires multiple samples per pixel, ala deep images. But it does seem like an obvious win, so it would be nice if more renderers picked it up (and Fusion added deep image support).

Technically speaking, before an image is recorded by a real camera, has to have its photons travel through a camera lens. So the image has a set DOF dependent on the lens aperture before it hits the camera sensor (or film). Then, a motion blur is created depending on the camera's shutter speed.

So, in real life it is DOF first and Motion Blur second.

But just because this is how it happens in reality it doesn't mean that's how you should do it Fusion.

Whatever works best!

Try both combos and if you see any difference stay with what you like. If you don't see any difference then none of the viewers will as well.

About the OPTICAL FLOW FEATURE:
hmm.. I didn't get that.. what can you do with this ?

the optical flow tool will try to find out how the pixels move between frames and build a motion vector from this. with this vectors you can then create retimed/motion blurred version of the footage

from a technical point the problem between DOF and MB cant be managed right:
in one way your depth blured image wont fit the motion vectors anymore
in the other way the you motion blured image wont fit the depth information anymore

I bump this thread because I'm trying to render a 3d composition with both MB and DoF activated and it seems they dont work fine together.

Separately, both effects render correctly. But if I swith on both, the result is very poor, looks like a 2-3 samples DoF.

I've set up a example scene, with strong perspective for DoF and a quick shake in camera to get motion blur.

Here are the test images:

1 Rendered with no effect.


2 with Depth of Field only activated, OK


3 with Motion Blur only activated, OK


4 Both MB and Dof activated: FAIL.



Two questions:

Is this a known issue?

Any fast and easy workaround using Fusion Free (no license for optical flow tools).

Thanks!

for working out layering of FX I would tend to coinsider how it works inside the body of the camera - starting from what happens when light first touches the front of lense to last which is what happens on the sensor (maybe even what happens from the transfer of data to the recording hardware when matching compession artifacts).

ie. in your scenario DOF field is the result of the aperture and lens characteristics.thats at the very front of the camera so that is the first effect to the footage. then motion blur happens as a result of how long the image is exposed onto the sensor, at the back of the camera so thats applied second.

so that is what would be 'technically correct' but then does it look right? OR sometimes does it give the best overall result. If flipping it on its head works for you in your situation and more importantly gets approval from client, then by all means flip it.

Thans for the answer, Jonathan


I dont think that's the "phisically" correct render.

It's true that DoF is in some way prior to MB. You have a brurred image in your sensor due to lens behaviour (DoF), and then it's exposed along a time interval, and if it's moving, you get MB.

In real life cameras there's no problem with that. You get a twice blurred image, perfectly smooth, because in the real life there's no time segmentation. Time is analogic ( :

Another different thing is the CGI simulation of those effects. Both are based on addition and interpolate of samples. Theres something in the Fusion8 algorithms, some kind of interpolation or optimization, that ruins the trick when you have to combine samples along time (MB) with samples in space (DoF).

I've tested rendering DoF and then using the Motion Blur filter in Davinci Resolve. But that filter makes artifacts where you have strong paralax, and the project I'm involved has massive parallax. (Its a kind of 3D HUD with graphics, text, wireframe 3d objects a video layers. A nightmare for any post-render MB filter, even optical flow.

The other way could be rendering with MB and try to apply a decent Z-channel based DoF in post. But again, a lot of layers of motion blurred wireframes is a nightmare for any DoF filter, even with z-buffer.

The only HQ workarounds I guess are:

- Render with DoF by layers, then apply Resolve's MB to individual elements (much less parallax, or none for most elements, will work quite fine most of cases).

- Render with DoF at 240fps and then convert that to a motion-blurred 24fps clip, using frame blending. That would simulate a 360º aperture/10 samples motion-blur. With a script it might be possible to render only the needed frames of a 1200fps timeline, concentrating the samples around the current frame. That would simulate shorter exposures and control quality. And you can process the samples with a good frame interpolation filter before adding them all...

I'm thinking while I write... modern frame interpolation effects based on pixel tracking would be great motion-blur simulators, just implementing frame blending with bias adjust.

Obviously I'm taking the first way in this low-cost project. Render by layers and apply Resolve's MB before compositing. That will probably be enough.

Hope new versions of Fusion can just handle MB and DoF properly, even if it means multi-pass render and big rendertime increment.

Hope new versions of Fusion can just handle MB and DoF properly, even if it means multi-pass render and big rendertime increment.

That's what's happening now, and has been that way all along.

... oh, I'm afraid I'll have some problem with post-render MB filters applied to flashing elements, drawing lines, etc.

May be I'll render fusion layers with MotionBlur and then make final 3D composition in a different scene using DoF.

Hope new versions of Fusion can just handle MB and DoF properly, even if it means multi-pass render and big rendertime increment.

That's what's happening now, and has been that way all along.

It's happening now, multipass is the answer, but in Fusion has problems. I guess both DoF and MB algorithms have strong optimizations. They are really fast!! But for some reason those optimization conflict with each other if applied simultneously. That makes sense, it's like crossing two patterns, you get moiré ( ; It seems one blur must be calculated by brute force repeating multiple passes of the optimized-tricked other.

But this is strange... I'd bet that most of times you have to compose 3D objects with real image footage you have to deal with both MB and DoF. Very few real action shots have infinite DoF, or are shot at 15º shutter. How do the VFX guys deal with this in Fusion? I'm suprised with this, actually.

Hope new versions of Fusion can just handle MB and DoF properly, even if it means multi-pass render and big rendertime increment.

That's what's happening now, and has been that way all along.

It's happening now, multipass is the answer, but in Fusion has problems. I guess both DoF and MB algorithms have strong optimizations. They are really fast!! But for some reason those optimization conflict with each other if applied simultneously. That makes sense, it's like crossing two patterns, you get moiré ( ; It seems one blur must be calculated by brute force repeating multiple passes of the optimized-tricked other.

No, it's far simpler than that. Not really an optimization, just a change in the usage of the inputs. When the DoF and MB are active in the 3Rn, the .Quality is ignored. .ShutterAngle, .CenterBias, etc., are used, though. The samples aren't multiplied, it's just using the .AccumQuality slider to determine the total number. In some cases, that will be a 5 digit number.

Ther must be an optimization issue because if not, it would be easy to solve this. Just make that every DoF sample (in space moving camera lens) is taken from a slightly instant of time in animation. Or the opposite, if you want; every time offseted samle for MB can be taken from a slightly different lens position. And you are ready. I used that trick to get MB-DoF from 1997's MAX 1 scanline rendering. (burning render farm).


Anyway, who knows how Fusion works internally. Not me, that's sure ( :


But now my question is, then, how do the VFX guys handle this? Almost any moving camera shot to integrate 3d on it has BOTH DoF and Motion Blur. What do they do?

im one of those VFX guys and I told you how I correctly would approach the problem. the only time I find myself flipping the order (DOF after MB) is when I am using very low sample in MB to save render time (if the final result is acceptable).

Im fairly new to fusion but know enough about the general compositing process to suggest that perhaps the order is not the problem, but rather the way or manner in which you are applying each effect in and of themselves.
For example in the image references you sent of the individual effect application are not incorrect but the overall settings do not seem sufficient for optically true effects. and the combination result shows that the way you have added the two effects together has not correctly done so. So in my first round of bug testing that is where i would focus my efforts - how to correctly combine.

I am sorry but I dont have the fusion answers for you, and it always has just worked for me in other apps so I am not sure.

Can you upload your test?

Ther must be an optimization issue because if not, it would be easy to solve this. Just make that every DoF sample (in space moving camera lens) is taken from a slightly instant of time in animation.

Anyway, who knows how Fusion works internally. Not me, that's sure ( :

You can determine it empirically, and you are correct. That's how it works now. So you just need to use more samples.