Problem with VrayVelocity 1.5Sp3a + Nuke 5.1

[muoto]

could you save an EXR with your two layers ?

[instinct]

and make sure that max aint that far out of range and make sure "clamp" is off.

Regards,
Thorsten

[3sgteMKI]

I'm going to have to give the multichannel .exr a try, i've been getting similar results (in fusion with the ReelSmartMB plugin) to what noel_d3 posted.

I usually add a velocity pass in the render elements, look at the "max velocity in last frame" and set my max velocity a bit higher, and while the result is sometimes passable during playback, if you inspect a single frame things look strange.

[muoto]

Well...if someone finds a way to have the vectorblur in nuke work, let me know...i thought i had a correct workflow, but it seems not...

thanks !

[instinct]

See the first page of the thread. Seems to work perfectly fine.

Regards,
Thorsten

[muoto]

Ok indeed, it worked with using the EXR multilayer from, vray directly.
Thanks!

[instinct]

Okay. Here's feedback on the problematic files:

They are saved as linear. hence the weird 0.241. When read as linear instead of sRGB they show an offset of 0.5. Offsetting that to -0.5 pretty much does the job.
I assume the offset is caused by turning on clamping. I assume VRay does that automatically as negative values would be clamped otherwise and hence it is remapping to 0-1 i assume.


Regards,
Thorsten

[sv]

nice one Thorsten. Good info. This is why I would only use EXR for CG graphics....!

[muoto]

Do you know as well how to get the relight of nuke work with the vraynormals?

[sv]

the foundry web page has some downloadable scripts for the relight which were shown at the masterclass....

[noel_d3]

There is in fact two differents type of Motion vectors pass :

RevisionFX ReelSmart Vectors v3 specifications, Red and Green color channels hold direction vectors scaled by normalized intensity - the speed - of the motion.

• Red Channel: X scaled by normalized Intensity [0,1]
• Green Channel: Y scaled by normalized Intensity [0,1]
• Blue Channel: not used.

According to the RevisionFX SmoothKit specifications, Red and Green color channels hold normalized direction vectors, and the Blue one holds the normalized intensity - the speed - of the motion.

• Red Channel: X [0,1]
• Green Channel: Y [0,1]
• Blue Channel: Intensity [0,1]

I am not satisfied with V-Ray Velocity pass compatible with RSM Vectors because the blur doesn't pass from an object to another, it works only if it was the alpha in the Vpass as we see in this result:



In seeing this video I had the feelings that using RSM Smoothkit this bug didn't appear and pixels was blurring from object to another.
I'm trying to use Maya lm_2DMV_v2 shader or Renderman to try to render this Vpass compatible with Smoothkit.
If I get there I would tell you if it's better.

In the meantime you can amuse yourself by trying to blur the image by render a exr velocity from this scene max and post the result : max scene

Regards,

Léon

[muoto]

yeah i have seen them. they seem a little over complicated though...
but are vraynormals ready to be used is there a "multiply Y with -1" involved somewhere as with vrayvelocity ?

[instinct]

@Léon:

Did a quick test and this is as close as it gets with post motionblur.

The reason beeing very easy and cannot be overcome without rendering in multiple layers. When blurring forground objects a lot parts of it need to become transparent. But the image does not contain what is behind these objects of course, hence the weird ghosts (for example at the center of the pencil in this example).

On a second note post MotionBlur does not account for rotational Motionblur. Hence strong rotating or curved movements are not properly blurred and lead to artifacts.

On a technical note on the attached image. I simply set the Velocity Element to be clamping off, max of 100 and rendered straight to exr via the VRayVFB.
In nuke no additional multiply node or shuffling or offsetting is needed.

This is just a straight vectorblur with a multiplier of 20 and mode set to forward + alpha set to the alpha channel.

Regards,
Thorsten

[sv]

Thorsten; regarding rotational motionblur (and this is getting way way over the top and i'm sure a bit too complex for production, but just asking from a theoretical point) would it theoretically be possible to export the entire scene as obj sequences (or FBX's), slap a motionblur3d onto each of the imported geometry's tansformGeo nodes to get the 3d vectors to pass to an individual vector blur for each object, or just use the motionblur in the scanline renderer inside nuke?
Would you this way be able to control the motionblur on a per object basis that would be correct for each object and therefore be correct when comped together?

P.S. > It;s Monday morning and even writing this has hurt my brain! I also appreciate this is a bit off-topic now so sorry for that!

[instinct]

In theory maybe. In reality not with a lot of issues and artifacts (non matching AA and alike prolly leading to all kind of oddies).

I have thought about a lot of ways to do proper rotational moblur (including some VERY weird ideas :P ) but so far i didnt even come up with a theoretical one that works properly.

The only real workaround would be rendering with higher FPS to get proper MoBlur by "retiming" the renderings. But then you can just as well render with moblur heh.

Regards,
Thorsten