Final Major Project – Post-production – Cholpat Saralamba's blog

UPDATE 01/11/2024:

After multiple iterations of the original version, I found myself dissatisfied with the story’s impact and the spectacle’s believability. This prompted me to explore alternate versions that could elevate the narrative. Through inspiration and references discovered online, I came across images that sparked new ideas. It intrigued me enough to begin reconstructing the story around this concept.

This alternate version will reuse the footage shot on set from the original version but will feature an entirely new background and storyline. It also replaces “Content A,” a previously scrapped project that was abandoned due to untrackable camera motion in the video.

The new version aims to surpass the original by introducing richer backstories, thrilling plot points, and elevated spectacles. These additions will create more tension and a stronger overall narrative, ensuring a more impressive and believable result.

Process of designing environment

After gathering references, I initially started designing the public park, where the story itself unfolds from. The park itself is based on a Chatuchak Park in Bangkok, Thailand. I sculpt the landscapes to closely replicate the real-life park, then randomly scattered trees and plants from Quixel Megascans to fill the spaces.

Then I moved on to the main “hero” building of the story, where the action unfolds into chaos. This building is also a real-life replica based on a currently under-construction “Mochit Complex”. I chose this building as the main facade of the story due to its halted construction which are left frozen in its place, unfinished. This reminds me of a similar incident that happened in 1997 when another skyscraper construction was halted amidst the Thailand economy crisis. That particular building had a reputation of being filled with homeless people, gangs and drug-traffickers. So I took that as an inspiration and apply it into this.

Layout Design

The premise of this story was to be styled like “found-footage” videos. So all the actions occurring during the storyline will be shot from ground-level by a group of influencers who happens to be there at the time this unfolds. I overlayed the first frame of the plate against the environment to try and get a perspective of what it would be like with the actors in-frame, without having to wait until compositing.

Camera tracking / export

This part is quite tricky for me, I’ve tried multiple softwares to try and track the on-set footage I’ve shot including, After Effects, Blender, Nuke and Fusion. But to no avail due to the fact that the movement of the camera is too extreme and unpredictable by automatic tracking. I’ve finally settled on doing it manually in PFtrack, an old-fashion industry standard software, alongside 3DEqualizer. But I chose PFtrack because of a more user friendly interface. So after masking out the actor movement in the plate and finished tracking and solving. I ran into another problem. Exporting into FBX and importing into Unreal does not work properly. I’ve spent countless days and hours into figuring it out, to the point where I almost gave up. However, I’ve found a tutorial online on this particular issue. Which involves using Maya as an intermediary between the two softwares (PFtrack and Unreal). Since Unreal does not understand ASCII FBX exported directly from PFtrack. I need to use Maya to bake the camera path and re-export as FBX 2020 that Unreal can understand.

Tracking in PFtrack

A tutorial mentioned on how to bake animation in Maya.

After the tracking data successfully imported, I can use the real footage as an overlay on top of the sequencer to see as reference when positioning the camera.

Building Design / Modelling

For modelling the buildings, I used a very mixed methods catering to different buildings and depending on how close it is to a shot. If for hero buildings (the one closer to the camera and needs fidelity) I will model myself and for some complex elements like cranes or detailed construction equipments, I will use KitBash3D assets to help blend in for more realism. I used variety of software for different purpose models. For distant buildings, I often used SketchUp, as it works great extracting models from satelite images and can be very convincingly real when looked at from faraway. However, sketchup exported models cannot be used directly in Unreal. So I had to use intermediary software like 3DS Max to import and reexport the mesh to be supported by Unreal

For more complex models and signage, I used blender as it supports SVG import (for signs and logos).

This model I’ve purchased from Evermotion

https://evermotion.org/shop/show_product/28-skyscraper-/13832

Characters

I have acquired an asset “Spec Ops” from BigMediumSmall as the model is very high detailed and comes with some useful preset animations. However, most of the animations I ended up using in the final sequence were from Mixamo.

https://www.bigmediumsmall.com/specialops

Crowds Simulation

For background characters, I opted to use Anima for its crowd simulation ability, the reason I used Anima over other solutions is for its fidelity in the realism of crowd models and animation. Bonus point is that it can be directly integrated into Unreal scene itself without having to export separate pass. It is a paid subscription service though. But it can be justified by how streamlined it was to my workflow.

When I finished blocking and planning out the scene in anima, I can export these as 3D mesh to reposition and adjust sizing, animation or settings inside Unreal directly. However, I could not preview the animation directly inside Unreal. So I have to render out the scene in low quality to check if the walking people overlap with any object or other people, which is a bit inconvenient when I want to make adjustment and have to render the whole thing just to check the animation.

FX/Simulation

I used Embergen as an alternative to Houdini as it offers real-time visualization of the simulation. I need smoke as a VDB format as I can position it in 3D space inside Unreal Engine.

Compositing

After rendering all the footage from Unreal. Which took around 3 days to render with 2048 samples and 3000+ frames. The reason it takes a lot longer than traditional Unreal renders is that I opted to use “path-tracing” which uses the same light ray tracing techniques as other offline renderers (Blender, Maya, Cinema 4D). I use Fusion to composite the Unreal plate and real photo plate together.

Final Node Tree

Sound Design and Additional ADR Session

After all the compositing and color grading was finalised. I arranged a ADR session with the original actress in the film to record new lines as the story was significantly altered from the original version