Curved motion blur for FLIP waterfall simulation

[gowthaman]

Hello everyone,

I am currently working on a waterfall Simulation using FLIP in houdini. I am using Geometry velocity blur to get the motion blur which is giving line streaks in render. How to achieve curved motion blur? I tried using deformation motion blur with more Geo Time samples but it is not working. Any suggestions?

Regards,

Gowthaman Ilango

M.F.A Graduate Student

Digital Production Arts

Clemson University

[tmdag]

geometry velocity motion blur is a vector that is being read on every frame.

You can get curved motion blur by letting mantra to calculate motion blur on render time. Turn off geometry velocity blur on object level, add Geo Time Samples on mantra node (eg. 10).

Check if you have date inbetween frames. if not, you could use timeblend to create it

[gowthaman]

Thanks for your reply. I tried your solution but it is not working. I want to make sure I am doing the process correct.

1. I saved the fluid surface as bgeo files in the disk.

2. Then imported the geo data through file node and connected the timeblend node to the file node.

3. Then set the Geo Time samples to 10 in mantra node and camera shutter time to 0.75.

What am I doing wrong?

[tmdag]

with fluid surface it might be little bit trickier as it need to track its motion somehow and for that you need a geometry that is not changing topology(pointcout) on every frame or at least to be constant in between frames so you could use shutter less than 1

[gowthaman]

ok, then what is the correct way to do without using fluid surface? Sorry, I am not an expert in houdini.

[tmdag]

it is not an easy task in that case. You could for eg. interpolate points before meshing and later on advect mesh with its velocities just on subframes.

In many cases of rendering waterfall, there is no even need of meshing points, maybe you could get different approach if curved velocities are so important for you.

[lukeiamyourfather]

To get proper motion blur on fluid simulations (with sub-frame curvature) you'd need to cache many states of the fluid simulation. For example caching every substep of the simulation instead of whole frames. Each sub-frame might have to be rendered independently and then combined to give the illusion of motion blur, basically what renderers used to do for motion blur. It would take a lot of disk space and render time. Mantra might be clever enough to figure out how to handle the multiple samples but I'm thinking it's based on consistent point count and topology.

[lukeiamyourfather]

You could for eg. interpolate points before meshing and later on advect mesh with its velocities just on subframes.

That could work too, sounds like a fun thing to try and implement. Could be useful in a lot of situations.

[gowthaman]

Thanks for your replies. I also emailed Mr. Jason Iversen regarding this issue and he suggested the following solution.

Curved motion blur with particles is rather tricky, as you can't do it with Velocity Attribute blur as there is only a single velocity represented in that info when you need multiple 'slices'; you have to synthesize some constant-topology shapes so that you can use Geo Time Samples (which is Deformation Blur, essentially).

You can do it with some clever SOP tricks where you analyse the particles over the shutter time and reject (or extrapolate) particle which are born or disappear in that time window, and then ensure you have a constant point count for that window, based on the "id" attribute, and then use Geo Time Samples instead of Velocity Attribute blur.

I am trying to implement this solution but finding it hard to figure out how to do this. Any suggestions?

[tmdag]

This is what i wrote on my second post but problem with your setup is that you are not using points to render but calculated geometry.

Still, to calculate point position of your points over the shutter time, you need to move it somehow and that is the tricky thing. You can't use velocities as it would be the same as using point velocity motion blur, and you can't compare frames in-between as they have totally different topology.

With rendering points itself (not recalculated topology) is easier as it is possible that at least you can track their IDs over time of their life

Edited October 30, 2013 by tmdag

[gowthaman]

But doesn't FLIP converts particles to surface geometry before rendering? If I am rendering just the points how can I get it to look like a fluid?

[tmdag]

it all depends on your shot if it is just a waterfall, rendering points might be enough and maybe adding geometry just in the centre.