Class 13: Refining & Denoising

This weekend, I decided I was going to start the process of denoising/sampling the opening shots and the first part of my sequence. I started reading the Redshift documentation and found out that there are 3 denoising engines that Redshift works with. I found that I got the best result with Optix, which is linked to NVIDIA.

I think these shots are completely done (except for some lens flares/color grading in post) but I'm really happy with the denoising that I got with Optix, so I decided to do a trial render of the first part using with Optix enabled. (**NOTE** This test has not been color managed, which is why it is washed out! These are the raw files straight from the farm)

This yielded pretty decent results (I had good luck with the farm for once), but revealed a lot of minor problems that I wanted to fix (before/afters posted below). Obviously, there's still some denoising/more research that I need to do regarding Optix and its usage on SCAD's render farm in particular (I don't know if it did much in the sequence or if it's wishful thinking), but I feel pretty good about wrapping up the first part by Wednesday.

I've also been trying to match my files so the cut between the files is seamless. I've gotten fairly close, but I need to go through with a fine tooth comb and make sure all of the lighting rigs have consistent placement/value in both files and that I'm getting identical results when rendering the same frame from each file.

Class 12: Troubleshooting Lighting, Still Texturing

I really wanted to fix the lighting iteration I had at midterms. I think I've done that, but I might've gone too far in the opposite direction. The issue did end up being the volume scattering, as well as light contribution from a couple of portals I have set up. I think moving forward I will try to implement some of the feedback with lighting that I got from my peers, like adding a few bounce cards where needed, or animating the volumetric effect.

Also while I was waiting for renders and whatnot, I reworked the parachute behavior so that it leans more upwards (correctly) rather than trailing behind the lego guy.

Besides tweaking some minor things in my simulation that I noticed in my render at midterms (car not moving all the way down the stairs, some clipping geometry, adjusting groups, recaching simulation parts to points rather than geo for more iterative freedom with UVs, etc.), I just kept texturing, mainly focusing on the Lego guys for now. I think the beginning frames in my midterm are pretty locked, so I'm going to begin the denoising/sampling process for those and try to piecemeal the renders rather than waiting all at once so the farm isn't a HUGE issue heading into finals. There's some stuff still that I would like to adjust (isn't there always!!!) but I think my iterative process is moving fairly quickly and I'm able to identify problems a lot faster than before.

Week 6: FX Refinement

This week I was focused on getting more iterations of the smoke/exhaust for shots 1 and 2, mainly getting critique from the mentors implemented in shot 1 and correctly setting up the render layers for shot 2 so the exhaust would show up better in post.

Here's a flipbook of the latest iteration inside Houdini. I think that the exhaust in the wheel well is a lot better than last week, however I think the behavior of the smoke still feels too wispy as it comes off the back of the car, so I need to experiment more in the coming week with noise in the disturbance/turbulence field (pulse and block size, amplitude, roughness, etc. in the pyro solver). This week I did add some fans on the back size of the car as well, as I saw that Deja has used that technique and gotten good results, so that might also help with the overall behavior.

Here are some videos I've been using as reference:

After troubleshooting with Eaza and Camilo we were able to get a render with the exhaust looking a little bit whiter/grey instead of the reddish-brown or black we were getting beforehand. Now that the render layers are properly set up, the FX can be color graded separately without affecting anything else in the scene. However, this color grade in particular reveals a lot of problems. With shot 010 the exhaust seems way too dense, and the attribute transfer radius on the ground is clearly visible here. There also is a mixing effect going on where the burnout trails look very smoke-like, while the burnout smoke around the tires still looks really dark, so this will need adjusting. Hopefully the solutions are as straightforward as tweaking the AiStandardVolume shader parameters. I also think there's some strange behavior right before the car takes off on the table; it feels like the smoke should almost wrap around the table instead of moving straight out. Right now, I am using a simple grid as the ground plane for the collisions, so I think this coming week I will also try to add some proxy geometry that aligns with the table in our scene and see if that affects the smoke's behavior at all.

Now that the exhaust is visible in shot 020, I would like to adjust the behavior. I'm in a limbo state with the effect right now of it either being too "poofy" in appearance or too erratic and choppy. Right now, I am using a particle system that emits particles from the exhaust pipes as the source for the pyro, so maybe the approach needs to change for this in order to achieve the look we want.