Motion blur on grain?

[DavidW]

I met a good friend of mine for lunch yesterday. He is also a shake compositor and works at one of the big four studios in London for the moment.
We came in on the subject of comping of course, and he mentioned a technique for making grain look even more photo real: They actually motion blurred the film grain they put on the CG comps. They actually took the camera motion and put a very subtle motion blur on the grain with a referenced Move2D-node...
This was totally new to me and I just wonder whether it's used a lot at different studios around the globe?
Other comments and feedback on the issue?

[djr33]

Hm. Interesting.

Grain is just random in each frame, not related to motion. So you wouldn't need to worry about any patterns, etc., in that sense.

However, if the camera is moving and everything is blurry, I can see that the grain would blur as well, though it's hard to say exactly how since it does seem to be sorta laid on top of the image, just pixels that didn't expose properly. If it's moving fast enough, I guess these badly exposed regions would move too?

I'll have to think about it some more (and hear back from others).

Perhaps some tests would help too. (Not CG, though those could be nice, but real shots with blur and not.)


How is grain generated? IS it pixels that go wrong on the CCD(s)? Or is it some property of light?
That's really what this comes down to.

No matter the motion, specific pixels wouldn't spread. But light rays would.

There's also the interpolation to consider, such as with deinterlacing (interpreting two frames and blending, so the grain pixels would blend), downsizing (either digital or in the camera, such as with a large resolution CCD), etc.

[DavidW]

Thanks for you comment djr33, but you seem to be talking more of video noise, and I don't know what approach that's the best there really...
I'm talking about pure 2K scans of film with no filtering/translation/interpolation at all that we comp on, and the matching grain that should be added to photo real 3D elements comped in.
The grain on real film happens in slight variation of light sensitivity in the actual celluloids of the physical film when exposed. I'm just a bit puzzled to this "motion blur on the grain" idea and if it's logical in any sense, or just a good way of simulating a natural phenomenon...
Other thoughts? :)

[djr33]

Hm. Film grain is random per frame as well, and, though organic, also based on particles, not on an overall effect. It's roughly the same as pixels (though it might take the space of several pixels in a frame).

It is also an effect like water on the lens, where it isn't really related to anything going on in the shot or camera motion.


I think it simply is wrong to add motion blur to the noise in a purely theoretical sense, but might help to match it to what happens unnaturally along the process with (as mentioned above) downsizing, resampling, interpolation, etc.

The motion blur seems like it wouldn't do too much, since it isn't actually moving. It's just sitting there. I wonder how different it would look to have just blurry grain. That makes a lot more sense than motion blur, to me.

[ShakeandBake]

It has always been my understanding that grain something that was a result of the chemical processes use to develop it. Therefor it nothing that was captured by the lens and wouldn't have any motion blur.


This may be a case of 'even thought it is accurate, it doesn't look good visually,' so they are taking a creative license on it.

[anthony113]

i agree with the above posters. grain has nothing to do with the motion in the shot or even the motion of the camera filming it.

according to wikipedia, film grain is the "random optical texture of processed photographic film due to the presence of small grains of a metallic silver developed from silver halide that have received enough photons".

it's something inherent to the film itself so therefore it wouldn't blur. it may be the case that in some circumstances it may look more "realistic" if the grain has a slight blur on it. but the way i see it, blurred grain would make the composited footage look less realistic as the rest of the grain in the original shot isn't blurred.

i've always added grain at the bottom of the node trees after all color corrections and everything, masked off from the original footage.

[Diogo Girondi]

Grain is part of the film emulsion and it has nothing to do with focus, camera movement, motion blur, etc. The only things that will make it more or less noticible is the exposure settings while shooting, negative type, development process, best light among other processes down the pipe.


There is no reson to add motion blur to grain, and you should always leave to add it in the end of the pipe while in comp since any operation like color corrections, blurs, mb, etc will affect the way syntetic or practical grain behave. Also film grain is more noticible in shadows and midtones and barely (if) visible in highlights, that's why is always a good idea to 'mask' your grain by the luminance of the shot (element).

[djr33]

Screen it onto a shot, I say.

As the last few posts confirm, motion blur is not applicable to grain.

However, if it makes the composites feel more right, then that's worth considering.

I think it's a random chance that it happens to be like this and the motion blur simulates a realistic effect.

Meaning-- what is it that the motion blur does to make it seem reasonable?

A few possibilities come to mind:
1. Areas with motion blur may simply feel better with less grain. So, blurring it makes it seem like less.. less noticeable, etc.
2. In the same sense, maybe just blurring it (as I mentioned above) would give a more subtle effect. (simple blur, not directional)
3. With both of these, it seems like maybe the better option is just have less noise, so the feel of it is the same as less strong noise.
4. The idea of lightness may come into play as well. If an element has the grain moving with it, then that makes sense... it's darker. So the real answer is to use screen, so the darker areas are affected, not the bright ones.

I still wouldn't mind seeing some tests.


There's also a chance that the grain being blurred worked well because it blended in with the element, giving it a more textured feel, or maybe like there was dust around it. Probably unlikely, though.

[Hugh]

Adding motion blur to the grain to make it sit in better sounds more like they've not set up their grain properly.


Sometimes, though, you need to do some really nasty things with grain to get it to actually work in a shot. If you're adding something like camera-shake onto a shot, often you'll be losing the grain that was in the original plate, as well as potentially the grain that you've added yourself. This isn't accurate, but it's one time when accuracy is just something that you may need to sacrifice.....

[mboden]

Adding motion blur to the grain to make it sit in better sounds more like they've not set up their grain properly.


Sometimes, though, you need to do some really nasty things with grain to get it to actually work in a shot. If you're adding something like camera-shake onto a shot, often you'll be losing the grain that was in the original plate, as well as potentially the grain that you've added yourself. This isn't accurate, but it's one time when accuracy is just something that you may need to sacrifice.....

What Hugh said... sounds right to me.

[DavidW]

Thanks for all the comments and discussions! :)
I've never done this myself (I always do the pure grain in the end of the comp in log space). I was just as puzzled when I heard about it as you guys so I haven't got any test images or anything. Just an idea of a technique that was in use...

[mondochiba]

I noticed that a lot of grain simulator lack a softness slider, real grain indeed tends to look softer and more blurred than many artificial grains do.

I sometimes simply generate a grain-plate by adding grain to a grey solid (using the original plate as reference), so it is easy to add that extra blur.

[Aruna]

I met a good friend of mine for lunch yesterday. He is also a shake compositor and works at one of the big four studios in London for the moment.
We came in on the subject of comping of course, and he mentioned a technique for making grain look even more photo real: They actually motion blurred the film grain they put on the CG comps. They actually took the camera motion and put a very subtle motion blur on the grain with a referenced Move2D-node...
This was totally new to me and I just wonder whether it's used a lot at different studios around the globe?
Other comments and feedback on the issue?

Woah, woah, woah.. Woah.. I've never done that. It's just not right. Sure, if you're doing a bit of post-processing on the plate, maybe via stabilization or slight transforms you could get away with it, but usually when doing something like that you should use a filtering method that retains the sharpness of the original grain.

Like everyone else is saying, grain is part of the film process itself, not a part of the original acquisition, so camera motions and zooms and pans have no effect on the blurriness or streakiness of the grain.

oh, just a quick edit. for camera shake and the like, the best bet is to apply grain to only the translating, motionblurred frames, and transition to the original when the camera shake is still. That way your grain doesn't overlap.

[nathan]

What are you talking about Aruna? Just look at the following case study.

4607 4608

Now tell me you shouldn't be blurring the grain. Just look at the different in them two comparisons and tell me its wrong. Look at how real the blurring makes the cg element fit into the comp. It's night and day I tell you! :D

[lewis]

I wonder if it might have helped the quality of the motion blur in the CG renders? Perhaps low quality renders with aliasing in the motion blur could benefit from some random detail being "combed" in the direction of the blur? Or some excessively soft CG blur, which could use some definition along that blur axis?

That would be a seperate issue to grain matching, really though.

I think you need to interrogate your friend further, David, to keep nrgy from losing his mind completely ;)

[Pnagle]

What are you talking about Aruna? Just look at the following case study.

4607 4608

Now tell me you shouldn't be blurring the grain. Just look at the different in them two comparisons and tell me its wrong. Look at how real the blurring makes the cg element fit into the comp. It's night and day I tell you! :D

is this a joke?

[Hugh]

is this a joke?

I don't think so.... I think he's making a very good point....

[€HR¶§†ØþHÊR]

I'm going to have a try at this, as nrgy demonstrated, its a very powerful technique.

[graemaster]

Hahaha...grain..the misteries of film...i love my job...

yeh david i suppose a defocus on the grain might achieve the same desired softness?? what you think!!? At the end of the day it is all just an illusion hey?

[hiphopcr]

I'll agree with the previous posters... motion blur on grain is weird. No matter what happens on a plate, the grain should be laid over the top and not be distorted or blurred, regardless of what is happening in the actual shot.

[unxetas]

well, we're almost starting to establish which of the big four does that, I'm pretty sure I've never seen anyone do it around here.. Weird, very weird.

One of the best compositors I've met while I was doing lots and lots of roto told me to ALWAYS respect the grain and you'll be fine. Adding motion blur to grain is not a form of respect as far as I'm concerned ;)

[hevonen]

I vote for sticky because this thread has pictures of cats.

Blurring grain (unless trying to simulate projector defocus or for matching purposes) is strange.

[PaddyW]

Film has the structure set in production of the material and the director of photography "deside" by lighting how much is visible. :)
If in the DI chain the picture moves just a little bit without getting noticeable defocused, there could be a blurred out grain...but in single frames not in shots!

So please tell me what is more noticeable:
The frame with the blurred grain or the blue-jeans worn in Gladiator?

[Gentle Fury]

Hmmm, thats just bizarre. Grain has nothing to do with motion and it wouldn't be blurred....ever. If I do a comp with cg elements grain comes last so anything that is motion blurred or defocused has the same grain as everything else.

Furnace has an awesome degrain node that you can actually use to sample the film grain to later apply to your cg elements, but i have never heard of blurring grain....it just makes no sense.