As mentioned in another post, I recently finished off an old animation project that I started working on (and almost finished) back in 2007. The animation is a music video wherein little paper cut-out versions of the members of Lothar perform their song Malone. The animation was made almost entirely using Blender and I thought I might share some information and thoughts about the process in case anyone is interested.

All modelling and animation was done in Blender 2.45. In terms of modelling, there is nothing too exciting going on; the action takes place in a simple box shaped scene. To achieve a convincing paper cut-out look, all textures were hand drawn on graph paper and then scanned. I played around with some different lighting options but finally settled on using only ambient occlusion lighting. After trying some different settings, I decided to go with a relatively low number (5) of ambient occlusion samples to get a slightly noisy look. Visually important objects, like the musicians, were emphasised by the use of materials with a small non-zero emission value. This made the objects a bit brighter than the surrounding geometry and ensured that they did not receive too much of the ambient occlusion shadows.


Getting the little paper musicians to play in sync with the music was a crucial part of making them come alive. Technically, this turned out to be a bit of a challenge, especially since the song was not recorded at a fixed tempo. To get the timing information needed for the animations, I recorded MIDI notes for each animation channel (drummer foot, drummer left arm, etc) by tapping the corresponding rhythm on a keyboard. The MIDI data was then converted to blender ipo curves using the midi importer script by julesd. Apart from manually recording the timing information, this process was fairly painless.

Another animation challenge was handling camera switching in some sensible manner. I used two cameras, one for moving and one for fixed shots. I alternated between the two by toggling a constraint aligning the fixed camera with the moving one. One benefit of this method is that no separate video editing step is needed so different shots and cuts could be tested more efficiently. However, this way of switching between cameras ruled caused problems with vector blur, that I initially planned to make use of. The reason is that the previous frame is used to compute direction vectors during the vector pass which introduces single frames of weird looking when switching beween the cameras.  If anyone has any wisdom to share on this topic, I’m all ears.

The final render was output as a sequence of PNG images, which allowed easy re-rendering of sequences that needed tweaking (there were quite a few). Each 1920×1080 frame took around 3 minutes to render on my old Core2Duo laptop. With a total of around 5000 frames, the whole animation, weighing in at around 7 GB, took a bit over a week to render. Since most of the objects are static, it may have been possible to reduce the render times significantly by baking the ambient occlusion shadows. I’ll probably look into that option if I’m doing a similar project again.

For post processing I set up a separate Blender 2.53 project to be able to use the new animation system that allows animating compositing nodes. The node setup is fairly straightforward and is used to add a vignette, a slight flicker, a bit of blur, some noise and a bit of colour curve adjustments.