Architectural experiments (for Blender Conference) with Cycles now with making-of

Some experiments I did using among others Cell Fracture, softbodies and Skin modifiers. I may finish it and polish a little when I have time. It’s the last example I used during my conference for the Blender Conference 2012:

And the updated flythrough:


ooooh O_O quite nice.

hey bro, that’s awesome, nice work after all hard work and time leads to a good modelling.

Thanks for the comments.
Here are some other shots of the flythrough I’m currently rendering:

And a NPR render, 100% cycles, fast and dirty (2 passes only):


And some new NPR renders. It’s a pitty it doesn’t work well for animation, because it’s so fast to render, and I’m quite happy with the result:

And here is the flythrough:

The camera movement are not so fine, but it’s still a wip.

I updated the flythrough, with two solar cycles, and some basic video editing. I must admit video editing is not something I know, so any advice for making it better is very welcome.
I still have to improve the noise on some parts, and correct some SunPosition bugs.

Man, that looks super awesome. How many passes? And also did you use some tricks to get better convergence in the interior parts or is just pure sun+sky? I agree the camera movement could be a bit more fluid, also the DOF is a bit too shallow which then gives the impression it’s a miniature.

I did a quick HowTo on my site: h

You will find the Blend file there too.
A video tutorial is made too, but I need to upload it on vimeo, which qill take a lot of time :smiley:

The video of the making-of is ready:

Thanks for this. I was quite impressed seeing that you use 2000 samples for most of the time. How long was the render and on which machine?

In the Blend file, I don’t remember how the samples were set. I realized during my work that I can animate the samples number. But before that I just rendered parts of the animation with different samples. I’m not so sure to remember but for the exterior shots, 200 samples were more than enough, inside with the big blur, 700 samples were not enough but i used it because of the time (around 20min/frame), and the fact that to realy get rid of the noise I needed more than 9000 samples.
I use a i7 with 6 cores with two gtx 580. So during day time only one gtx580 was calculating. And at night both of them.

I did new tries and even 4000 samples are not enough to get rid of the noise… I think it’s time to use node editor to remove those remaining noises…

Maybe change the Clamp value a little…