Post-Production – Into the Houdini-Verse

From the beginning of my project, I had planned to use Houdini, but the amount of Houdini work has significantly decreased. I realized that some of my ideas, such as wall destruction, didn’t align with my focus for the project or my career goals. I wanted to dedicate most of my time to Nuke, and while there would still be extensive compositing work for anything I did in Houdini, it would take away time from Nuke. Additionally, as my project was shortened overall, the heavy Houdini work I initially wanted to do was removed anyway.

However, I recognized that Houdini still had a role to play in my project, particularly for the portal shot where leaves are scattered around the area. I needed to find a way to incorporate environment interaction with the leaves, but it didn’t start off well.

I searched for suitable leaves to duplicate and simulate, but the best leaves I found were scanned at a high resolution, making them inefficient for simulation due to their mesh complexity. Despite my patience in waiting for Houdini to calculate (and often crashing in the process), it was clear that this approach wasn’t feasible.

Furthermore, Houdini wasn’t yet available on the render farm. Taking matters into my own hands, I repeatedly requested its inclusion and even contributed to some Python code that made it possible. Ironically, I later discovered that I wouldn’t end up using Houdini’s renderer anyway.

Nevertheless, I needed to get something working in Houdini, and the simulation proved to be more challenging than I initially anticipated. Eventually, I found a solution to the issue of high-resolution leaves by procedurally creating a simpler leaf mesh and using high-quality textures I found online to compensate for the lack of detail (since they would mostly appear out of focus or motion blurred).

First I created a Geo node that would host the actual leaf creation, the labs collection for Houdini by SideFX already has a HDA that was perfect for this.

I then created a small grid, randomized an orient attribute per point in the grid and created an attributewrangle for randomizing the scale of the points of this grid.

Then with a copy to points and a copy node for that copy, I was very quickly able to create a few slightly different iterations of the one initial leaf that was generated.

I created two of these, one slightly larger than the other depending on where I wanted these leaves to be placed, the process that follows was the same for both, however.

I took those objects, packed them, and created a simple rigid body simulation of them in DOPs.

All the leaves fell to the ground due to gravity, creating a natural looking scatter of the leaves that would be tedious to replicate manually. Because of the way Houdini works, these leaves are incredibly unstable due to how thin they are, meaning the collision geometry causes them to constantly be bouncing, to mitigate this I first render out the final frame as its own geo.

I then took that back into another DOPs operation and made a similar setup to before, only this time I changed the collision type to prevent the bouncing (also key framing when it would be active). Then, in addition to the standard gravity system, I added a fan, set to replicate the timings of the portal appearing on screen.

The result of this simulation can be seen here:

Unfortunately, even after succeeding in getting Houdini on the render farm, its renderer Mantra kept giving me undesired results. Even if the leaves could be graded and composited to look better, the camera seemed to not track in the way I wanted it to. So I decided to export the leaves and take them into Maya to render using Arnold, I ended up duplicating and offsetting some of the leaves because it didn’t feel like there was enough initially. That, combined with the HDRI produced a result I was much happier with, even if it wasn’t perfect.

However, whilst I had found a leaf texture, I realized the colour of it didn’t really match the other leaves in the scene, I originally thought a bit of variety would be realistic, but it simply didn’t work. So using a few nodes I was able to adjust the softness, colour, and add motion blur, and with everything else added to the comp, whilst not being perfect, it seemed to work fairly well.