Victor Perez

Visual Effects Supervisor

The Beauty of Imperfection: Human Craft in the Age of AI

Abstract: As AI makes visual beauty instantly accessible, cinema enters a new creative era where production value and artistic value begin to separate. When images can be generated by anyone, the central question is no longer what can be shown, but why it should be shown, and, just as importantly, what should remain unseen. This talk explores how AI is shifting visual storytelling beyond spectacle and technical achievement toward meaning, authorship, restraint, and human imperfection. With a focus on cinematography and visual effects, it argues that the future of the image will not be defined by flawless beauty, but by craft: the deliberate act of shaping, selecting, withholding, and piecing beauty together into emotional and narrative meaning. Maybe we needed AI to remind us what makes us human.

Biography: Victor Perez is a renowned figure in the global visual effects industry. His career spans over 28 years so far, and he has worked in many areas of visual storytelling. Starting as a photographer and cinematographer, extending his interest to the whole image post-production process and evolving into a visual effects artist with a special focus on photorealistic compositing of CGI. Along the way, he has worked on a number of Hollywood films, including Christopher Nolan’s ‘The Dark Knight Rises’, ‘Rogue One: A Star Wars Story’, ‘Harry Potter and the Deathly Hallows’, and ‘Pirates of the Caribbean: On Stranger Tides’, plus many more. Eventually, he became a Film Director, and his career naturally led him to his current role as a Visual Effects Supervisor. What really drives Perez is his focus on storytelling. He is fascinated by how new technologies can help filmmakers tell stories in ways never seen before. His work is a fine alchemy of Science, Technology, and Art, reflecting his passion for innovation and creativity.

Niloy Mitra

University College London and Adobe Research

Neurosymbolic Structure for Controllable Visual Media

Biography: Niloy J. Mitra leads the Smart Geometry Processing group in the Department of Computer Science at University College London and the Adobe Research London Lab. He received his Ph.D. from Stanford University under the guidance of Leonidas Guibas. His research develops machine learning frameworks for generating high-quality geometric and functional content in computer graphics applications. He has received several recognitions, including the ACM SIGGRAPH Significant New Researcher Award (2013), the BCS Roger Needham Award (2015), and the Eurographics Outstanding Technical Contributions Award (2019). He was elected a Eurographics Fellow in 2021, served as Technical Papers Chair for SIGGRAPH in 2022, and was inducted into the SIGGRAPH Academy in 2023. Beyond research, Niloy is an avid DIYer and enjoys reading, cricket, and cooking. More information is available at https://geometry.cs.ucl.ac.uk .

Robin Kahlow

Runway

Simulating the Stack

Abstract: Every pipeline that produces images, sound or interactive experiences is built from layers of explicit systems. Drawing on Runway's research into world models and real-time simulation, this keynote maps which of those layers learned simulation can handle today, where explicit systems still hold and what comes next.

Biography: Robin Kahlow is a Principal Research Scientist at Runway, building the real-time technology behind Runway Characters, Solaris and GWM Worlds, which generate conversational characters, interactive interfaces and explorable worlds. He led the research on Gen-Turbo, which brought high-quality video generation down from minutes to seconds, and helped build Runway's Gen series of image and video models, from pretraining to inference optimization.

Stuart Penn

VFX Supervisor at Framestore · Production keynote

Project Hail Mary: A VFX Journey

Abstract: Grounded in practical production, the visual effects for Project Hail Mary were built on a close collaboration with the filmmakers spanning from early pre-production through final delivery. Virtual Production was an essential tool for planning the shoot effectively. We will explore how Rocky was brought to life: a seamless integration of digital assets with an on-set practical puppet. A significant technical challenge was maintaining the transparent xenonite barrier separating Rocky and Grace. Executing the Zero-G sequences required digitally reconstructing costumes and sets to remove stunt rigs, while preserving the actor's original performance.

Biography: Stuart Penn is a Visual Effects Supervisor at Framestore in London, which he joined in 2000 after a decade as a research physicist. His scientific and artistic expertise has supported acclaimed productions including The Dark Knight, Guardians of the Galaxy, Gravity, Paddington 2 and Avengers: Endgame. For Oscar-winning Gravity, he helped develop the techniques used to shoot its zero-gravity scenes, working with motion-control robots and the film's innovative LED lightbox. Having led Framestore's work on Bong Joon-ho's Mickey 17, he has put his scientific acumen to use on Phil Lord and Chris Miller's adaptation of Andy Weir's Project Hail Mary.

Daniele Siragusano

FilmLight

How to train your Looks - Towards a Differentiable Colour Grading Pipeline

Abstract: A motion-picture colour pipeline is a chain of image-processing stages assembled over several decades. Much of that chain predates the requirement of differentiability, and predates the availability of any smooth, expressive model of the creative look; looks were therefore captured empirically, by measuring film stocks, fitting colour models, constructing colour spaces and appearance models with branches and clamps, and freezing the result into a three-dimensional look-up table. The resulting pipeline is only marginally differentiable, and this property, rather than any absence of suitable methods, is the principal reason that machine learning, which depends on end-to-end gradients, has made limited progress in colour grading. We report an end-to-end differentiable colour pipeline. Its input transform is a linear matrix; its display rendering transform is a genuinely differentiable transform, reported separately; and its creative look stage, historically the final and most difficult component, is rendered differentiable by Chromogen, a smooth parametric look model, in combination with Chromogen ML, a method that recovers Chromogen parameters from an arbitrary existing look, including the baked look-up tables that encode the photochemical film look. The baked and the staged pipelines produce visually indistinguishable images, yet only the staged pipeline admits a gradient, and it is this property that restores the applicability of machine learning to colour grading.

Biography: Daniele Siragusano is an image engineer at FilmLight. He leads the design of Truelight Colour Spaces, Baselight's colour management system. He also created the scene-referred grading tools BaseGrade, X Grade and Chromogen, and the Truelight CAM display rendering. His current work covers ML-driven grading, spectral imaging, film grain and visual adaptation in HDR.

Russ Harding

Partner and Director, Rhino Rock Studios

From Screen to Game to Screen: How Immersive Is Becoming the Bridge

Abstract: A single story world can now move between cinema, console, headset and phone. Drawing on a career spanning PlayStation's EyeToy and EyePet, Wonderbook: Book of Spells, and immersive experiences including Peaky Blinders: The King's Ransom, Infinite Inside and Project Hail Mary, Russ Harding explores how the camera learned to see the player and how film and TV worlds became playable. Looking ahead, he considers how real-time capture, AI and the latest headsets could carry audiences from screen to screen. The talk closes with an open discussion: who authors a world that lives everywhere, and how could research and production build that bridge together?

Biography: Russ Harding is Partner and Director at Rhino Rock Studios, developing original IP, games and interactive experiences in collaboration with entertainment partners, alongside co-development and outsourcing services. With more than three decades in interactive entertainment, he has helped pioneer camera-based and augmented reality play, from PlayStation's EyeToy and EyePet to Wonderbook: Book of Spells, created in collaboration with J.K. Rowling. His credits also include Doctor Who: The Edge of Time, Peaky Blinders: The King's Ransom, Thief VR: Legacy of Shadow, the mixed reality adventure Infinite Inside, and Project Hail Mary: Journey Among the Stars. Through Rhino Rock Studios, Russ also advises studios and IP owners on interactive strategy, narrative and immersive production, and welcomes conversations about collaboration.

He Sun

Milk Visual Effects and Royal Holloway, University of London

From Technical Capability to Team Structure: Real-Time VFX in the Post-Production of High-End Fiction Television

Abstract: Real-time game-engine technologies are moving from previsualisation and on-set virtual production (VP) into post-production visual effects (VFX) pipelines. Existing organisational research on VP concentrates on work at or before principal photography, and little is known about how a real-time team changes the division of labour in post-production, where most VFX labour remains. This paper reports a practitioner case study of Good Omens, Season 3 (2026) at Milk VFX, where a centralised five-artist Unreal Engine (UE) team and a conventional departmental pipeline operated concurrently over one shot list, with each shot routed according to its technical requirements. Drawing on production tracking, shot-level workflow breakdowns and the delivered artefacts, the paper maps each deployed real-time capability to the pipeline dependency it altered and to the organisational decision taken in response. Three findings emerge. First, the Unreal team absorbed asset, environment, animation, FX and lighting for the shots it owned. Second, the relationship between the Unreal and Compositing teams shifted from sequential hand-off to concurrent work on shared material. Third, specialist departments were retained, not replaced, for work the engine could not hold at delivery quality, so the facility ran two organisational forms over a single shot list. The paper argues that real-time capability is best treated as an input to organisational design, and that departmental specialisation becomes a deliberate choice rather than a default.