Machine-learning denoising in feature film production
Henrik Dahlberg, David Adler, Jeremy Newlin
DOI: 10.1145/3306307.3328150

One Context. Every AI.

Marrow gives people one place to control the work, preferences, relationships, and routines that their AI tools use. Each app or agent receives only the context relevant to its task, with citations, explicit evidence gaps, and approved corrections that remain available when the user changes tools.
I am an engineer and consultant specialising in AI and computer graphics. At Industrial Light & Magic, Metaphysic, and Unity, I have developed rendering tools, trained models, and operated systems on film sets and live stages.
I contributed to two Academy Sci-Tech Award-winning technologies, Disney's ML Denoiser (2025) and ILM's Lama shading system (2026), both distributed through Pixar's RenderMan. As engineering lead for ILM's AI denoising initiative, I scoped, evaluated, and integrated a denoiser used on over 100 films since 2018, including every Pixar film since Toy Story 4. My Lama contributions included layerable shaders for ABBA Voyage, plus procedural noise, dispersion rendering, and hair shading shipped on Star Wars: The Rise of Skywalker, No Time To Die, and Marvel's The Eternals.
At Metaphysic, I built and deployed real-time AI face, voice, and translation systems. I developed the neural voice translation pipeline and face swap suite for Eminem's 2023-2024 campaign, operated live systems across 80+ Drake arena shows, built live AI voice translation for Cai Guo-Qiang's TED Talk, and delivered voice cloning and holographic face swap technology for the COP28 opening ceremony in Dubai.
Earlier work includes real-time rendering for ILM's StageCraft virtual production system at Pinewood Studios, shipped features for The Mandalorian and Obi-Wan Kenobi, and real-time path tracing at Unity Technologies. I hold an M.Sc. in Applied and Computational Mathematics from Aalto University and KTH Royal Institute of Technology, with specialisations in Mathematical Statistics and Light Transport Simulation.

AI systems for live performances. Rendering technology for film.

AI voice and face work across Eminem's 2023-2024 campaign. Built the neural voice translation pipeline and real-time face swap suite for Houdini, PR campaigns with Complex, and the 2024 VMAs live performance.

ILM digitally recreated the four members of ABBA as photoreal digital humans for the Voyage concert. Four continuous performances were synthesised from the original band members and stand-in actors. I worked on rendering technology and on-set support, and developed the layerable shaders used for all character and costume rendering on the project.

Live generative AI visuals for Drake's North American arena tour. I served as machine learning engineer and live operator across 80+ shows across North America. Responsible for face swap data curation, model training and deployment, on-site machine setup, system maintenance, client interaction, and real-time operation during each performance.

I built and deployed the real-time Mandarin-to-English voice translation pipeline for artist Cai Guo-Qiang's TED Talk. A custom voice model reproduced Cai's own voice in English, preserving his tonality, timbre, and cadence. The translated speech was output through the PA system as he spoke on stage. A first in TED history.

Rendering technology and procedural terrain generation using Phasor Noise, a procedural noise function implemented as part of the Lama framework.

StageCraft virtual production rendering via the Helios real-time engine for the Disney+ series. On-set technology operations at ILM's LED volume stage.

Developer on Helios, ILM's real-time render engine (Vulkan Ray Tracing) for the StageCraft LED volume. Shipped ray-traced refraction, parallel texture compression, and existence mipmapping with alpha distribution.

First film with full Lama shading deployment and first ILM production to use the ML denoiser on photorealistic live-action content. I worked on rendering technology and lighting for the CG Genie character and environments.

Voice cloning and live face swap pipeline recreating the late Sheikh Zayed bin Sultan Al Nahyan as a hologram for the COP28 opening in Dubai, UAE. The holographic Sheikh Zayed walked on stage and delivered a speech in Arabic to assembled world leaders.

AI face and voice synthesis for the Still Gin commercial by Dr. Dre and Snoop Dogg. The concept featured their interaction with AI-recreated Frank Sinatra and Sammy Davis Jr. in a Rat Pack-era jazz club. We worked with the Sinatra and Davis estates to ensure faithful representation.

Rendering and digital artistry for the Netflix nature documentary narrated by Morgan Freeman. Photorealistic CGI creatures spanning 4 billion years of evolution.

Implemented game mechanics and artist tools in Unreal Engine 4/C++ for THQ Nordic's open-world action RPG.
Industrial Light & Magic · 2018-2021

ILM's layered material shader framework, distributed with Pixar's RenderMan. A modular, physically plausible approach to material composition that expanded shading workflows at ILM and led to broader industry adoption via RenderMan. Since 2019, all ILM CG assets use Lama.
Industrial Light & Magic · 2018-2021

Engineering lead for ILM's AI denoising initiative. Owned scoping, research evaluation, production testing, and pipeline integration of a U-Net denoiser for Monte Carlo path tracing, developed jointly with Pixar, Walt Disney Animation Studios, and Disney Research. Contributed to pipeline optimisation that enabled large-scale production use and the commercial release in RenderMan 25. Used on over 100 films since 2018, including every Pixar film since Toy Story 4.
Industrial Light & Magic · 2021-2023

Real-time cinematic render engine built with Vulkan Ray Tracing for ILM's StageCraft LED volume virtual production system. Shipped production-facing rendering features and supported delivery in real on-set conditions at Pinewood Studios. Used on The Mandalorian, Obi-Wan Kenobi, and other productions.
Metaphysic · August 2023

Presented real-time face swapping technology live on stage at SIGGRAPH 2023 Real-Time Live! in Los Angeles.

The Academy's Sci-Tech Awards video features Industrial Light & Magic's Lama team receiving a Technical Achievement Award.

Lama receives a Technical Achievement Award for its layered material shading framework, now used across the industry via Pixar's RenderMan.

Disney's ML Denoiser receives a Scientific and Engineering Achievement Award. Used on over 100 films since 2018.

Brahma acquires Metaphysic at a $1.43B post-transaction valuation. Metaphysic's voice and face-swap technology becomes part of Brahma's enterprise AI platform.

First-ever VMA awarded for a digital character created using an AI-powered pipeline.
Henrik Dahlberg, David Adler, Jeremy Newlin
DOI: 10.1145/3306307.3328150
Xianyao Zhang, Marco Manzi, Thijs Vogels, Henrik Dahlberg, Markus Gross, Marios Papas
DOI: 10.1111/cgf.14337
Henrik Dahlberg
US Patent 12,373,909
Presenter · Los Angeles