Lego Fallingwater

The Lego Architecture series has intrigued me since it first appeared, and this past Christmas I received a set of my own.  I thought it'd be fun to document the build process brick-by-brick.  A YouTube search revealed that my idea was not original by any means (over 50 similar videos!) but I still thought it'd be fun.  I've been wanting to try some stop motion with my older Sony digital camera anyway, and this seemed like as good a subject as any.

I built a camera stand from Legos, lacking a small enough tripod for the height at which I planned to shoot.  The whole setup was simple, and admittedly crude, and I would make a few changes if I were to repeat this process.  For one, the camera stand needed to be slightly more rigid, as evidenced by the slight jitter between frames.  For another, although the Sony image quality looks good it requires two AA batteries to run and this film went through two sets of batteries.  In the future I'll be getting an HD webcam if I decide to produce another similar film.

The "Studio" Setup

Although the stop-motion build-up was fun, I really enjoyed adding the live waterfall to the scene.  I used Blender with the Blam add-on for static camera matching.  The whole matching process was very straightforward once I got the latest version of the add-on installed.  (NOTE: when upgrading an add-on from an older to a newer version it's necessary to remove the old version, install the new version, AND restart Blender.)

I don't think I got the Lego geometry, for channeling the fluid simulation, exactly right.  It looks pretty good in most spots.  The 3D view of the geometry is shown below.  I tried to constrain the flow as it left the Lego creek bed.  This worked for the first part of the simulation, but eventually it overflowed and ran out the sides.

3D geometry matching the Lego creek bed (roughly)

To overcome this "leakage" problem I ended up creating a mask and compositing the Fallingwater image over the top of the fluid simulation sections that I wanted to hide.  I know where the masking is happening, and it doesn't look particularly obvious to me.  Hopefully that's true for the casual viewer as well.

This was the node setup used to combine the two shots and mask the overflow.

The Toddler Strikes Back

Creating my "Human Cloning Works" video was highly educational for me.  I learned a lot of new stuff about Blender's compositor and rotoscoping.  I wanted to keep that trend going, so I found another post-production challenge: light sabers.  I admit that these kinds of videos are a dime a dozen on YouTube.  I doubt any other subject matter could be as cliche for aspiring amateur effects artists.  Probably because they're not too hard to do, and they can be fun to watch.

I knew how to mask out the saber blades from previous experience, but I wasn't sure how to get the saber effect.  I experimented a little, but nothing looked like what I wanted.  Eventually Google turned up an example for me that I tweaked a bit to arrive at this node setup:

The mask is on the left, and gets split into three different blur nodes.  The Fast Gaussian gives you the larger aura, with the other two lesser blurs completing the inner glow of the blade (in two layers).  The color curves can be adjusted to whatever blade color you want.  This whole thing is then overlaid on the original footage with an AlphaOver node.

I made four masks in all.  One for each blade, one for any foreground body parts that needed to obscure the blades during the fight, and one for the wall that blocks part of the view towards the end.  The overlay order went like this:

• Original footage
• Blue blade
• Pink blade
• Wall mask
• Foreground mask

The only other detail I added was to brighten everything with a Gamma node (at the output of the original footage) any time the sabers collided.  It's crazy convenient to be able to add keyframes to anything, including node control values.  I thought about getting more complex with this concept, like masking the characters or certain parts of the room and just brightening them based on the location of the fight during any given clash.  I lost interest in this idea, but I'm still pretty pleased with the added complexity added by the simple brightness tweak.

I noted on the cloning post that I did the audio mix in Audacity, and a helpful poster at BlenderArtists informed me that the Video Sequence Editor could have been used.  I decided to give it a shot this time.  I still feel pretty uninformed about the ins-and-outs of the VSE, but I did figure out how to cross-fade a few segments, and control segment clipping and fade-in/out effects for both audio and video clips.  Here's my final strip setup:

The VSE was really indispensable for doing the audio looping on this clip.  I originally envisioned figuring out the time codes where I wanted all of the saber sound effects, and then putting it all together in Audacity.  That would have taken a long time, and the results would definitely have been sub-par.  Using the VSE I could drop the effects right on the time line, in multiple tracks, and check the sync in real time.  It was so simple to tweak and adjust the timing.  I'm definitely a believer in the VSE, even if I still have a lot to learn.

So, with no further ado, here is the finished product:

And now, for my next trick, I think it's time to try a little motion tracking.  We'll see how that goes.

Cloning

After watching Tom Jelfs' "Blend That Film" tutorial:

 ... I decided to try out this technique myself.  I was using Blender 2.61 at the time.  Following Tom's instructions was pretty straightforward.  I wanted to expand on his ideas and branch out a little bit, so instead of creating twins I went for triplets.

Here's the final compositor node layout:

The only real variation between what Tom did and my layout is that he only had one mask to deal with, which he put into a renderlayer which could be used directly in the compositor.  I had three sets of overlaid footage, requiring two masks.  I figured out how to put each mask on a different renderlayer, but only the active renderlayer was updated in the compositor.  There may be a way to make this work correctly, but I didn't find one.  My workaround was to render the masks out into two image sequences, and then feed the mask files back into the compositor.  It worked great, but it did add an extra render step.

There were a couple of lessons learned in this process:

• The lighting between my three takes, even though they were filmed consecutively with minimal interruption, was different enough to be seen in the final composite footage.  I attempted some manual color balancing to minimize the effect, but there was no way to do that effectively for the whole film.  In the future, if I was really concerned about matching the shots, I would be more conscientious about controlling the lighting (i.e. no natural daylight).
• The leather couch in the shot changed its form enough between takes that it was noticeable in the final result.  I would pay closer attention to the setting, and eliminate hard-to-control elements like leather upholstery if they might create inconsistencies between shots.
• I came across a handy application, the AoA Audio Extractor, when finishing the soundtrack.  This software allowed me to pull the audio track off each of my three original clips.  I then dropped each track into an Audacity project and created a single audio file to merge with the frame sequence output by Blender.  The audio and video was combined using VirtualDubMod.

Even though some things could have been done better, I was still very pleased with the final film.  See for yourself!