Motion Blur alternatives?

[Craiglet]

Anybody know of any tricks or alternatives for good motion blur on fast moving objects? I am not a fan of LW's motion blur-- I can spot that 'stutter step' look a mile away...

I know that you can dither the blur, which helps to a degree, but still leaves something to be desired in my opinion. Cranking up the AA settings helps a bit as well, but multiplies my render times exponentially...

I've heard of a plugin called Vector Blur. I don't know much about it but I believe its a post process effect and have heard it adds a huge render hit as well. All my shots render out as image sequences, so I wouldn't be opposed to a photoshop or after effects process as long as it creates the desired effect: natural looking motion blur! =)

Anyone have any juicy tricks or tidbits on this topic?

[TyVole]

People have told me that ReelSmart Motion Blur, a plugin for Premiere and all major compositing apps, is 10x better LW's. DFX+ also has a motion blur option.

In my opinion -- by its very nature -- it would be better to apply motion blur in post.

[Mylenium]

That's not exactly true. ReelSmart Motion Blur is quite good, but only with small amounts of blur and not so fast movement. since it does its calculations based on differences between frames it also produces glitches quite often (unwanted blurring in certain areas, wrong vectors) - rotating objects are a good example here.

Mylenium

[TyVole]

Do you think DFX+'s motion blur suffers from the same problem?

I don't have ReelSmart, but your comments are honestly the first negative ones I've heard; and I've read at least a half-dozen reviews.

[HowardM]

uh heard of it? its part of LW, why not just try it out?
Vector blur doesnt really add much of a hit, and its great!

[Mylenium]

Hi everybody,

Don't get me wrong - ReelSmart Motion Blur is good, in many areas its quality even beats LW's Vector Blur (it doesn't screw up the alpha for instance) but it never can beat true 3D motion blur. It's the same like with adding DOF etc. in post - since it is all done on the "baked" image it needs to re-calculate blur vectors etc. depending on color depth, resolution and so on this may lead to miscalculations which in turn can produce glitches. Critical areas are for instance the outer borders or overlapping objects that move in contrary directions.

Final conclusion: RSMB can help sometimes, just don't expect miracles

Mylenium

[jin choung]

tips-

render your anims in PASSES if you're gonna do post motion blur.

vastly reduces artifacting from FG/BG foul ups.

also, be wary of DFX+ motion blur. if it does not do pixel analysis, it may essentially be the same as the one in AE... that is, it gives motion blurs to things that you keyframe motion to - NOT the motion of pixels in an unmoving layer. i don't know but be wary... personally, i doubt it.

and howdy again mylenium... so we meet again :)

again, since motion blur is an artifact of image acquisition (like barrel distortion and lens flare - and it does not exist in physical reality), i still don't get your argument for a 'real 3d' motion blur....

jin

[kevman3d]

Vector Blur is OK - Though it does occasionally create artifacts on the edges of the blurred items. However that said, its fantastic for really fast motion however, like race cars zipping past the camera, etc.

There's also two options in Vector blur - The second 'Blur' will actually blur your native LW AA motion blur passes quite nicely (though occasionally with artifacting as well, so try it first).

If you plan on using DOF, you can also apply 'Digital Confusion' as it respects AA passes and can actually also improve the Motion Blur - However it seems to be based on similiar code as Vector Blur (once it crashed with an error caused by 'Vector.p' - Hence the conclusion!), and there are a lot of artifacting issues making it not too useful for Animation work anyway... Sadly... :(

For a short film project we're working on, we bumped AA up to Extreme to get a nice blur on film-res plates. Anything lower gave us way too many staggered pixels! :)

BTW, the freeware 'Virtual Dub' also does Motion blur - Though it does look pretty funky sometimes (sorta dreamy in a wierd kinda way). Thought I'd just mention that, but I won't recommend it as a good solution! :)

[Edbittner]

If you are considering doing blur in post, check out this test for an "Akira-motorcycle-trail" test I did.(Strictly Lightwave,low AA).
Ed

[ericsmith]

again, since motion blur is an artifact of image acquisition (like barrel distortion and lens flare - and it does not exist in physical reality), i still don't get your argument for a 'real 3d' motion blur....

While this argument makes a certain degree of sense, it doesn't really hold up in practial application. think about this extreme example-- If you were to set a camera up in a room and lock the shutter open for a minute or so, and then walk around in the frame, then the end result would be a highly transperent blurred image of you all over the place. The key point is that the background wouldn't change, and would not be blurred. The opacity level of you would be a number basically derived from the amount of time you were at a specific location devided by the overall time of the exposure. This is what 3d motion blur is really all about. Sampling the moving object at it's various locations throughout the time of exposure (shutter speed) and adjusting the opacity of each sample based on the number of samples (number of anti-alias passes) taken. If you do motion blur in post, the computer will never know what is behind the object in motion, and will have to interpolate data, simply bluring the edge of the moving object with the background next to it.

the real problem with any post process is that they really break down when there is a lot of this missing information. Reelsmart works well when the motion is slow enough to give the plugin lots of information and little to interpolate, but when you have an object going fast enough to move an inch a frame, it has no way of figuring out all the data that is missing to fill in the motion blur.

Eric

[ericsmith]

If I can sum up my point it would be this:

in 3d, without motion blur you are basically taking a snapshot with a shutter speed time of zero. This is of course impossible in the real world, but due to the infinte nature of time, you are technically taking an infinite number of exposures and blending them together onto the film. So in reality, a real camera does the exact same thing as 3d motion blur in lightwave, but with an infinite nubmer of AA passes. (that's why you don't notice the stepping artifacts in a real life photograph.)

Eric

[Mylenium]

Hi Jin,

I understand what you are trying to say, but ther is still good reason for 3D MBlur. You see, blurring is not the problem but the direction (vectors) of the blur are the problem. These are normally lost after rendering so any 2D post algorithm needs to re-calculate them. Since it does not know about object coverage or certain other phenomenons there is a good chance it will produce wrong results. Also, like someone stated, in real life you have theoretically infinite oversampling (it is only limited by the sensibility and size of the emulsion) which will correctly reproduce curved motion blur. In rendering this is often lost even with 3D motion blur since the number of passes is limited. In 2D this is even worse since all vectors are transformed from 3D world space to 2D screen space. So there you are. BTW, if you wanna talk about phenomenons that are camera or perception related you could go on for eons arguing about lens flares, color shifting and clamping, lens distortions, chromatic abberation, dispersion up to Einstein's theories on photon warping. I don't see much sense in this argumentation even though it is scientifically correct and may be of some importance to few people. I feel it's better to leave out such considerations when only a quick and simple solution is asked for.

Mylenium

[jin choung]

i only bring up the artifacts to differentiate between things that are present in 3d space (reality) and things which are simply elements of image acquisition.

as for the blurring the background, this is why i recommend doing the motion blur on the f.g. separate from b.g. - in passes. indeed, the bg shouldn't be blurred.

but yah, the vectors argument works!

you've sold me mylenium....

jin