tl;dr: Humain Studios faced a 33-year bottleneck cleaning 44,000 scans from advanced 4D scanners. Instead of scaling manual work, they built an AI-driven pipeline powered by Dell Pro Max workstations and NVIDIA RTX PRO GPUs. This shift enables rapid NPC creation for AAA titles like Halo and Call of Duty, transforming character workflows and unlocking creative freedom.
Humain Studios has built a reputation as the go-to partner for Microsoft’s Xbox Studios, creating everything from populated airports in Flight Simulator to the facial rigs that bring beloved characters to life. As CEO and founder, Greg Maguire brings decades of experience in visual effects work at Lucasfilm and Disney. The studio’s expertise in facial action coding systems comes from its Chief Scientific Officer, Dr. Erica Rosenberg, who oversees the scientific framework for describing every possible human expression. Their work spans AAA titles including Halo Infinite, Call of Duty, and the cinematics for Diablo 2 Resurrected, with a team that’s grown to 26 people over eight years.
The Belfast-based studio has perfected an impressive pipeline. “We can put a character, like an NPC character together in less than a day. Sometimes, we build about 200 NPCs a night,” Greg explains.
A new collaboration at Studio Ulster
Humain Studios has enjoyed top talent and a close collaboration with Studio Ulster, a fellow Belfast video production facility. The $100 million facility features the world’s most sophisticated VP stages. Most importantly for Humain, it houses three state-of-the-art scanners that represent a massive leap forward in performance capture capabilities.
Through their collaboration with Studio Ulster, Humain Studios will operate and manage this unprecedented scanner setup: a 4D scanner capable of capturing facial expressions at 60 frames per second, combined with a 3D scanner and a full-body scanner. With 60 full-motion video cameras and the ability to take very high-resolution imagery every 10 frames, it’s exactly the kind of breakthrough technology that pushes character creation to entirely new levels.
But this opportunity presented a nearly impossible problem.
The 33-year problem
Greg did the math on what this new scanning capability would generate, and the numbers were staggering. His team had previously spent a full year with 10 people cleaning up 4,000 scans. The new 4D scanner at Studio Ulster would generate approximately 44,000 scans every year. “I estimated that the data that we would need would take us 33 years to clean up,” Greg realized. The traditional approach of manually cleaning scan data simply wouldn’t scale for this volume.
Instead of trying to speed up manual processes, Greg and his R&D team decided to automate them. They developed machine learning algorithms to handle the scan cleanup to detecting and removing unwanted elements, segmenting the data, and preparing clean surfaces ready for character creation.
“We looked at machine learning for the past five years to automate parts of our pipeline,” Greg explains. “But then we kind of figured out that instead of replacing parts of our pipeline, we could replace our entire pipeline.”
The approach required training sophisticated AI models on massive datasets, running complex 3D point cloud processing, and handling multiple data modalities simultaneously. But this kind of workload demanded substantial computational resources, particularly memory capacity for large batch sizes and high-performance GPUs for training.
The hardware that made it possible
Robert, Humain’s head of R&D, had experimented with various hardware configurations over the years, from desktop GPUs to remote cloud servers. Nothing provided the combination of power and accessibility they needed. Remote servers created bottlenecks with their massive datasets, while existing hardware couldn’t handle the memory demands of their machine learning workflows.
The breakthrough came with a Dell workstation powered by an NVIDIA RTX 6000 Ada Generation Graphics Card. Equipped with powerful graphics, training jobs that previously took three days could now be completed in approximately six hours. More importantly, its substantial memory capacity allowed them to dramatically increase batch sizes.
Having powerful hardware on-premises solved another crucial challenge: working directly with their massive datasets without the delays and complications caused by uploading and downloading hundreds of gigabytes of scan data.
Unleashing creative confidence
The performance boost fundamentally changed how the team approached their work. “You are more brave sometimes with experimentation,” Robert reflects. “You think, okay, I can squeeze even more out of it. I’ll just do one more experiment. It really encourages you to keep going.”
This confidence to experiment has been crucial as they develop their automated scan cleanup technology. Their machine learning tools now handle the complex process of detecting unwanted hair and artifacts, segmenting scan data, and using prior knowledge from their clean data to guide the automation. They can train models that work across different modalities, project learned annotations from 2D to 3D and process point cloud data directly.
The result is a system that can automatically clean scan data which would have taken their team years to process manually, turning the 33-year problem into a manageable workflow.
The future of character creation
What started as an impossible data processing challenge has evolved into something much bigger. Humain Studios is now pioneering a completely new approach to character creation. The combination of advanced hardware and powerful on-premises computing makes a measurable difference. Dell workstations provide the performance and memory capacity needed to train and run the AI models that transform raw scan data into production-ready characters.
Building tomorrow’s stories
Humain Studios is expanding what’s possible in digital storytelling. Whether populating Flight Simulator with believable passengers and crew, creating the facial rigs that bring Halo characters to life, or developing the next generation of automated character creation tools, their goal remains to give creators the products they need to tell better stories.
The journey from a 33-year impossibility to automated reality proves that the biggest breakthroughs don’t come from working harder, but from working smarter. When creative vision meets the right computational power, problems that once seemed unsolvable become steppingstones to innovation.
