This week consisted of a lot of research about clouds. We needed to answer several questions in order to start accurately developing in Houdini.
Mapping the Shots
Our sequence has 4 shots total, and it was important to know where we were in the sky for each shot.
The location tells us which types of clouds we'll be seeing, as well as the ambient temperature (this is also important for the pyro sim). This week, we focused on stratocumulus clouds, which make up the main bed of clouds in the final shots. Considering our sequence takes place during sunset, the temperature will likely lean to the lower end.
Forming the Clouds
Stratocumulus clouds are the combination of the low, flat stratus clouds and the white, puffy cumulus clouds. They tend to form in lumpy rows and groups, which tend to be regular.
My FX partner (Joonho Bang) and I decided to divide and conquer. My first instinct when thinking about implementing in Houdini was to see what other people in the industry had done with the issue of CG clouds.
═ Provided by FX Mill mentor Todd Akita (Real Time Weather Clouds in HZD):
═ Two articles discussing clouds in The Aeronauts:
Based on these references, I thought modeling, then adding volume data to the created VDBs and shaders would be the best route. I decided to test the capabilities of this with a simple model.
Above is a basic sphere with a mountain node. I knew that Houdini had some cloud nodes already built in, so this basic geometry was fine to start with.
A cloud node and cloudnoise node.
Above is a compilation of some of my process for getting the right look with the new HDRi made by my teammate Jake Diana. As you can see, I changed the shape eventually. This was rendered with mantra using a billowy smoke shader. It was a combination of modifying the shader and the actual volume data inside to get a light scatter. (Clouds are made of water droplets and ice crystals, causing them to do this quite uniquely).
This was the final result of the model-method test.
Ultimately, we decided that we would go with the modeling method instead of the pryo simulation method that Joonho had been working with, since we had more control over the shape and finer details this way.
A Brief Look at Pyro
Though most of this week was cloud-centric, I took a bit of time at the end of this weekend to start studying the pyro simulation I'll be making for the gas burners. The main takeaways were as follows:
fueled by propane (no smoke)
average about 4-4.5 meters in height
velocity is dependent on the height (ruled by pressure of gas release)
I'm going to be using sparse pyro for the sake of time. I haven't done too much detailed work with fire in sparse, mainly smoke, but I attempted a little sim:
I ran into a strange issue that I am still troubleshooting where I'm not getting any visualization of the flame in the viewport or render, despite the fields being there, hence the white fire and black background. Right now, the flame needs more volume and velocity control.
I also noticed in the reference videos (see blog 3/29) that there is also sometimes a bit of flame lick near the burners from excess gas being lit by the coils. This may have to be its own separate simulation.
Final Notes
Unfortunately, our pipeline is still a little unclear. Both myself and our lighting artist (Bua Kanjanapongporn) are having issues installing Arnold in Houdini. Additionally, the SCAD renderfarm has been having issues with rendering Arnold render nodes from Houdini when used through the farm (not vlab). So, we've been considering other methods, but each brings challenges that we're still in the process of weighing.
For next week:
🖽 Establish pipeline
🖽 Start clouds for other shots
🖽 Start cloud layout and think about composition of each shot (get renders for Jake)
🖽 Continue R&D/lookdev on pyro
┄┄☁┄☁┄☁┄┄
コメント