Compositing and Composting

These words are not only similar orthographically, but bear some kind of other relationship too.

It’s sometimes necessary to apply the latter to the outcome of the former.

For me this ‘sometimes’ was even more frequent during my work on Ara’s Tale through the last days.

With a lot of iterations, tries and errors I managed to finalize the first 3 shots of this short. And when I say finalize, that doesn’t mean that they are never to be touched again, but that these shots are very near to the that version, I will call final ( in some not too distant future I hope …).

Even though very exhausting, I deem these last days as very valuable as learning experience in the area of lighting, render layer organization and finally compositing.

Each of these shots had its own unique difficulties and I had to discover them all and had to find solutions and sometimes workarounds for them.

Some of the problems encountered with the first shot ( the cage closeup) I already discussed in my previous posts, so I won’t go into any detail here.

Shot 2

For shot 2 I have a long shot where Ara walks across the bridge. It is night and yet I want to be able to see the environment. It turned out, that for these long shots it won’t be necessary to use SSS and thus hugely simplify the render setup. The most work here was to find a proper balance between Ara and the environment and to give everything a convincing depth. During my first tests I always got a – what I call – studio look, where you just didn’t buy that this shot is actually taking place in the open. Its still there but much better now.

An interesting problem surfaced with my kind of workflow I am using here.

I usually setup the lights and renderlayers and render 3 to 4 shots covering the whole shot sequence. I render these shots with 50% of the final resolution. If everything works ok, I turn on my render server and create a 50% resolution raw render sequence and feed it through the compositing network. With this I see the first time the animated result and can easily spot discontinuities in light and colors. These can be fixed, sometimes with animated values for the various control parameters.

This gives me a ‘final’ node setup and its time to render the shot in full resolution, feed it into the nodes and hopefully have a final shot in full resolution.

It turned out that the luminance of some of the glow effects are dramatically higher when going to higher resolutions. I have yet no idea why this is so, but this just adds another loop in my workflow iteration toward a final highres shot.

Shot 3 (see image on top)

Shot 3 is a medium shot on Ara. SSS was not to be neglected here and I have already discussed the issue with splitting lighting into renderlayers and SSS in this post.  I will post a detailed breakdown of this shot in the next post, so just a few highlights here.

Hair turned out to be a massive troublemaker in regards to rendertime. The need to split the SSS layers to extra scenes and the wish to have have full control over the individual lights during compositing led to a setup where the hair gets rendered five times for one frame. On my quadcore a 50% resolution frame now takes ~ 6 minutes to render going up to ~ 20 mins for a full res frame. I tried several approaches to reduce the time doing compromises but in the end returned to my original render setup. It takes time but it give me the best results.

Another nice coincidence was the announcement by Sebastian König of his compositing tutorial. I actually bought it and went through all the 5 hours of material. While not telling me something fundamentally new it was filled with various gems of bits of knowledge and just niceley expanded my current knowledge of this topic. For anyone wanting a more in depth look into the compositor should really consider this one.

And now after so much talk I proudly present a release candidate of the first 3 shots of Ara’s Tale

And finally something for the readers interested in some statistics:

current amount of hours spent: 919

1 hour of rendering on my renderserver (only one at the moment, so only 12+12HT cores) takes 0.3 kwh. Shot 3 took ~3.5 hours to render the 50% resolution version and thus used ~1kwh, which translates to ~0.18 € in the place where I live. I have attached a watt meter and currently the total consumption is at 17kwh.

Leave a Reply

Your email address will not be published.