Monday, October 01, 2007

The force-field streamlines lines plug-in that I disinterred from my vfx wannabe days inspired me to reprocess the fluid sim video with the velocity streamlines drawn in. I really like the result, and it suggests to me a host of variants. For instance, I would like to trace them very densely, or perhaps visualize the velocity magnitude as a heightfield with lighting effects.

I was surprised at the way the streamline lines squiggle furiously sometimes. And while it's evident that they are responding to my movements, they also seem to be weirdly autonomous.

I kept the number of streamline sparse in this one, since when I added more, the quicktime encoder produced absolutely dreadful results -- muddy, artifact-riddled, barely legible. The youtube version is actually only marginally less ghastly. I will post a decent quicktime later.

[Later]There are a couple quicktimes of the effect here.


Blogger Julian Butler said...

Holy crap dude this is awesome!!! MOAR please. Do you think it'd be possible to produce images from just the fluid trace lines given enough iterations etc? ie: recognisable forms like your face?

2:41 PM  
Blogger The Method Artist said...

Oh, hey, thanks a lot man! In this version, the sim is a wild beast that is hard to make do anything specific.

One thought I had was to drive the sim using vectors orthogonal to the gradient. So strong edges would have streamlines that closely followed them. It would be intensely stylized, but it might just be possible to make out a recognizable form.

2:49 PM  
Blogger Julian Butler said...

perhaps recognisable forms are a red-herring in the production of beautiful line-art... what about resolution? Is the input bitmap limited to webcam sizes or could you feed it a much larger single frame from a DSLR say?


5:32 PM  
Blogger The Method Artist said...

I totally agree that half the fun is letting it do its thing and seeing what you get. The computer is usually a better artist than me ;-)

For real-time, as coded, it had to be a small image -- 320x240.

The same code with minor tweaks will run at arbitrary sizes. It was quite slow on HD resolution, though -- several seconds per frame, typically.

The bottleneck is the fluid sim, which is a crap, single-threaded thing I wrote an age ago. There are a few freebie CUDA-based ones floating about which could conceivably give real-time HD performance, with a bit of engineering. 1024x768 would likely be no problem.

5:41 PM  
Blogger Julian Butler said...

aha... wow a few seconds... I was imagining the results on something at print res, like maybe 5k or so and turning on all the bells and whistles like antialiasing etc... although I guess at that size it'd not be necessary really...

I'm going to come and say hi at your desk tomorrow sometime... although I think you said you'll not be there? Either way... this stuff is great!


11:57 PM  
Blogger The Method Artist said...

Yeah, I'll be out of town until Monday, with sporadic email until then. But definitely drop me a line at my gmail address any time -- I'm kevin dot atkinson. It's great to be talking to other artists at Weta finally.

The limit on resolution is that it has to be video, because the optical flow is what drives the fluid sim. I'd love to try this again using a Canon Mark II or something. Or a Red camera, for that matter!

I saw you had a show not too long ago -- sorry I missed it. A ton of awesome work on your site.

12:58 AM  

Post a Comment

<< Home