The Safe Haven Project
I was given a shot for a project outside of school, I was asked as set supervisor on this project so I was pretty excited to start working on an actual shot for it. There was already some initial comp work done on the walls and on the construction in front of the camera, but the shot felt too empty to be in a war torn zone. So I was asked to add some debris and destruction to make the shot feel more full. It had to be done really quick so I wasn’t planning on wasting time and building everything from the ground up, I reused some assets from other projects that my classmates made and modified them to fit the shot.
Figure 1 This is the original shot, there has been some compositing work done on it by someone else but there was still a lot to add to make the environment feel more destroyed
The first layer of destruction to add to the scene are small parts of debris, something that gets scattered across the ground to suggest a larger world that’s been destroyed. I got some assets that my classmates made for another project that involved a destroyed city.
First of all, all the assets were in one obj file, so I had to divide all the assets into multiple primitives to scatter them randomly across the ground. I got the objects from a classmate but couldn’t get to the maya save file, so instead of annoying him with something he has nothing to do with I decided to split the models up my own way which probably isn’t the best way to handle the situation but it did quickly fix this issue. I simply converted the geometry into a high quality volume, then converted it back to a polygon soup and then used polyreduce to get the amount of polygons to a reasonable rate, this way all geometry that intersects with each other will be seen as one primitive. I used an assemble node to split them up and with a for-each I assigned each piece of debris to a random point that has been scattered across the ground. I lifted these points a bit up in the Y axis and dropped them into an rbd sim. Now the objects touch the ground and have random rotations. After caching these objects I split them up into concrete rubble and bricks, I created a point wrangle for both groups to control the percentage of how much is on the ground, this way I have control over the debris after placing them. I also put a simple grid on the ground with a paint sop, I paint a path and use attribute transfer to transfer the colors to the points and delete them based on the color, so in addition to reducing a random percentage of debris, I can also paint a path to remove the rubble, this way I have the most control over the layout of the debris without having to scatter them a second time.
The next part was to add some bigger assets that fills the screen. Again I took the liberty to reuse some of the assets that my classmates made for other projects, so I didn’t model them because I was on a tight schedule. I grabbed a truck from a project and tore out the glass. I wanted to try a physically accurate simulation to deform the car like it had crashed. But after using FEM and deforming rigid bodies, I decided I’d stop wasting time and just use a ripple SOP and a pointVOP for extra detail. I also added a jersey wall that the wall has crashed into. The texturing and modeling was already done on it, just like with the previous assets. I assigned the textures to them and rendered one frame in Houdini to Arnold. It took some tweaking, but eventually I found a way to render Ambient Occlusion and have a different pass for ground shadows to have more control in comp.
I took the frame into Nuke and composited it in, it was relatively easy and I only had some troubles getting the colors to match. This is the shot I sent to the director and he actually had another idea for the shot since the cars don’t really fit into the world. So I had to remove the cars and just duplicate the jersey walls. Since the assets has been made only once I had to add some variance to the new assets so it isn’t too obvious that the same asset has been made. The quick approach I found was to clip the asset in half and then mirror it. So I can have assets that are fractured at other places or not fractured at all. I did have to redo the UV’s and quickly make the same texture in substance painter. Eventually it works and I managed to extrapolate three different assets from one asset.
After that I wasn’t quite finished, the director liked the setup but wanted more destruction in the scene, I tried to add some rubble in the lower right corner but the perspective didn’t match. So I eventually downloaden some textures and multiplied them with the scene and with some of the CG assets to give them more variance. Like with the rubble on the ground, I didn’t want them to all be just grey, the bricks should have a brick like color and texture, same for the concrete rubble. I downloaded some decals from textures.com, multiplied them with the scene and then placed them with a 2D cornerpin node to get the right perspective. It wasn’t particularly difficult work it was just a lot of busy work. Eventually I ended up with a comp the director (and I) liked. I want to stress that I didn’t make all the assets nor did do all the comp work. You can see exactly what I did when you compare the first picture to the last and keep in mind that I didn’t create the assets, just modified them and scattered them.