SIGGRAPH Talks '25: Proceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Talks

Perlin noise is an integral tool for generating procedural textures for computer graphics applications. It’s widely used across multiple industries and applications due to its simplicity, speed and controllability. We present 2 simple changes to the Perlin noise algorithm that allows for the inclusion of anisotropy in the output noise. We believe these small adjustments open the door to new texturing workflows, as well as simple vector field visualizations in multiple dimensions.

We advertise the use of tetrahedral grids constructed via the longest edge bisection algorithm for rendering volumetric data with path tracing. The key benefits of such grids is two-fold. First, they provide a highly adaptive space-partitioning representation that limits the memory footprint of volumetric assets. Second, each (tetrahedral) cell has exactly 4 neighbors within the volume (one per face of each tetrahedron) or less at boundaries. We leverage these properties to devise optimized algorithms and data-structures to compute and path-trace adaptive tetrahedral grids on the GPU. In practice, our GPU implementation outperforms regular grids by up to × 30 and renders production assets in realtime at 32 samples per pixel.

A critical step in VFX production is capturing the movement of actors to integrate 3D digital assets into live-action footage. In recent years, advances in regression-based computer vision models such as human detection and motion models have enabled new workflows to emerge where parts of the Match-Animation process are automated. However, challenging shots that contain ambiguous visual cues, strong occlusions, or challenging appearances can cause automated systems to fail and users must revert to manual specification or to the previous generation of semi-automatic tools based on local feature-based tracking [Bregler et al. 2009; Sullivan et al. 2006].

Our key insight is that regression models can be used not only at the beginning of the process, but throughout by using manually specified cues. For example, given a partially detected actor, the user can specify a few landmarks manually, which once re-injected into a model, will yield new detections for the rest of the body. Based on this insight, we developed new tools that significantly reduces the time required for complex shots, combining automation with human expertise to overcome the limitations of current markerless motion capture systems.

Real-time feedback of character rigs is an important feature for every animator [Kahwaty et al. 2019]. While improvements to Maya’s execution engine has helped to run character rigs faster in parallel, it has been challenging to use high-end rigs directly in animation. Instead, proxy rigs are utilized to give animators instant feedback, but with the drawback of providing a low-quality preview. Deformations can also be approximated using a machine learning network, but this requires a time-consuming training process [Öhrström et al. 2021]. This paper describes how we have implemented complex deformers as OpenCL kernels that work with Maya’s GPU evaluator. We will explain how we have ported less GPU-friendly algorithms, such as BVH construction, to the GPU while achieving accurate results. This has drastically reduced the need to create proxy rigs and provides animators with an interactive user experience without sacrificing visual fidelity.

HNTB, a global architectural and infrastructure engineering firm, has partnered with Cesium, a leader in open and interoperable 3D geospatial platforms, to enhance the design, visualization, and contextual analysis of large-scale infrastructure projects. By adopting Cesium's technology, HNTB has significantly reduced the manual effort required to model complex projects—from small toll plazas to extensive 26-mile roadway alignments—by leveraging Cesium's 3D Tiles and geospatial streaming capabilities. To further streamline workflows, HNTB developed custom tools within Unity and Unreal Engine for editing Cesium 3D Tiles, including volumetric masking, edge blending, curvature adjustments, and the addition of natural elements like trees and foliage. Cesium supports AEC (Architecture, Engineering, and Construction) workflows by providing out-of-the-box geospatial context, such as Google Photorealistic 3D Tiles, and by optimizing and hosting user data for performant runtime streaming. Enhancing this ecosystem, Cesium continues to innovate with tools like the Cesium Design Tiler for efficient BIM model streaming and Cesium ion for Autodesk Revit, enabling seamless export and publication of Revit designs as 3D Tiles.

For the Look Development of The Wild Robot, characters needed to exhibit the richness, physicality, and motion of real-world fur and feathers. But unlike a realistic simulation, the look mimics how an artist applies detail to a painting: in layers of brushstrokes that form simplified groupings of light and color, with surgical placement of highlights. ‬While fur and feathers are traditionally shaded as curves, these features needed to respond to light like continuous surfaces to better fit into the stylized painterly world. To achieve the desired look within our traditional modeling, grooming and simulation workflows, new Look Development tools were developed including new shaders, custom groom attributes, an all-new feather system and new LookDev lighting rigs. Custom normals, UVs and derivatives were baked into the groom curves that provided a continuous canvas for a new shader that used artist-controlled ramps to stylize how these curves respond to light at different angles. The shader provided separate layering for each of the specular and diffuse lobes, written out to additional “accent” AOVs. While the final look came together in compositing, our renderer added a mechanism to preview this look with its built-in Display Filter compositor. ‬The resulting characters meshed beautifully with the stylized organic island sets, while maintaining the desired physicality.‬‬

Animating a bird’s wing fold in computer animation is a very complicated task. Even after achieving a beautifully folded pose, keeping the wings firmly on the torso during actions and performances often requires labor-intensive counter animations. This talk introduces the surface constraint wing fold system developed for The Wild Robot, enabling animators to maintain folded wings seamlessly during dynamic movements. By reducing manual adjustments, this system streamlined the process, saving time and allowing animators to focus on compelling performances.

Cinematic visual effects have long been the domain of high-end hardware, reserved for movie studios and powerful gaming rigs. But what if we could bring that same level of quality to mobile devices—without sacrificing performance? AI VFX was born out of this challenge: to push the boundaries of real-time visual effects on mobile by merging Generative Adversarial Networks (GANs)[Goodfellow et al. 2014] with the VFX Graph.

In this talk, we’ll take you behind the scenes of AI VFX, a technology that enables stunning, real-time fire and water effects that run seamlessly across the full spectrum of mobile devices—from six-year-old Android phones to the latest iPhones. By integrating AI-powered enhancements into the VFX Graph (GPU particle system), we’ve developed an approach that balances visual fidelity with efficiency, ensuring cinematic-quality effects can exist in mobile gaming, social media, and beyond.

We’ll dive into the technical breakthroughs that made this possible, including synthetic data generation, real-time inference optimization, and workflows for combining GAN[Goodfellow et al. 2014] and VFX Graph—all while overcoming the hardware constraints of mobile platforms. Through recorded demos and real-world examples from TikTok, you’ll see firsthand how AI is revolutionizing mobile VFX, unlocking new creative possibilities for developers, artists, and storytellers.

Disney Animation makes heavy use of Ptex [Burley and Lacewell 2008] across our assets [Burley et al. 2018], which required a new texture streaming pipeline for our new real-time ray tracer. Our goal was to create a scalable system which could provide a real-time, zero-stall experience to users at all times, even as the number of Ptex files expands into the tens of thousands. We cap the maximum size of the GPU cache to a relatively small footprint, and employ a fast LRU eviction scheme when we hit the limit.

In Walt Disney Animation Studios’ "Moana 2", the main characters undergo significant evolution from the original film, displaying a broad range of emotions while interacting with diverse environmental conditions. Crafting the hair and cloth motion to consistently support and enhance the complex character performances presented challenges for the character team, particularly when combined with nuanced art direction. To this end, we developed a multi-tiered and strategic approach for the choreography of the hair and cloth which includes three main areas of emphasis: Performance Categorization, Continuity through Visual Planning, and Iterative Refinement. Integral to this process were extensive and detailed drawovers, to ensure that hairstyles and costumes reacted consistently. This strategy allowed us to frontload the work through careful planning, enabling technical animation to execute on art direction of shots efficiently. We illustrate the overall approach with specifics for hair with Moana and Maui, and with Matangi for cloth.

Cloth draping is a prevalent tailoring process that gives 3D form to sewn 2D panels of fabric. However, when dressing animated characters, artists often prefer to model garments with delineated spatial structures and clean silhouettes at the cost of diminishing the presence of folds and wrinkles produced by draping. To reconcile stylization and realism, this work describes a new approach for directing cloth draping that accommodates 3D shaping and 2D pattern making simultaneously. Our key contribution is a method that generates custom UVs blending the distortion induced by 3D shapes into 2D fabric panels. As a result, we can retarget cloth simulations to compute physically plausible draping deformations that smoothly transition to prescribed 3D forms. To assist the garment design, we also propose a flattening tool that constructs low-resolution UV panels amenable to 2D manipulation. We showcase our results with a series of garment assets and cloth animations from Pixar feature films Inside Out 2 (2024) and Elio (2025).

This talk explores the integration of traditional Jacquard weaving techniques into our existing workflow at Netflix Animation Studios. We discuss the merits of fibre-level construction and the lessons learned as we laboured to place the power of a Jacquard loom into the hands of digital artists.

At MPC, we have recently transitioned from a proprietary, procedural solution to a new, non-procedural technology called Loma. Mufasa: The Lion King was one of the first projects chosen to use Loma in a production setting, primarily with the goal of creating higher fidelity grooms needed by the 20+ hero characters on the project without the performance limitations of a procedural system. However, non-procedural workflows come with their own limitations so, in this paper, we will discuss some of the problems we faced during production and the solutions we developed to navigate them.

This work emphasizes the environmental impact of gaming, highlighting the energy consumption of retail devices and the carbon footprint of gameplay across several titles, with a focus on the Call of Duty ® gaming franchise. It presents practical steps that real-time graphics developers can take to reduce energy consumption in a device-agnostic manner. Care is taken to ensure this effort does not have an adverse effect on gameplay standards or business requirements. Each step taken is backed with appropriate metrics, including feedback validation across the gaming retail landscape.

Machine-Guided Spatial Sensing is a novel measurement technique that combines augmented reality (AR), active learning, and human-in-the-loop interaction to measure environmental fields with high accuracy and efficiency. This system employs a head-mounted display (HMD) and a handheld sensor to capture various physical quantities, such as flow fields and gas concentrations, in real-time. A central data model processes the collected measurements continuously and updates predictions of the environmental field. Using active learning, the system identifies regions of high uncertainty and guides the operator to optimal sampling locations through intuitive AR visualizations. This closed-loop framework effectively transfers the sampling expertise from the operator to the machine learning algorithm, enabling efficient and accurate field estimation. Experimental evaluations demonstrate that the proposed method achieves high accuracy and reduces measurement times significantly compared to traditional sampling techniques. The system’s flexibility allows for integration with various environmental sensors, making it suitable for applications in engineering, scientific research, and environmental protection. By leveraging real-time data analysis and human-machine collaboration, Machine-Guided Spatial Sensing provides a robust, user-friendly solution for complex spatial measurement challenges. Future research will focus on enhancing sensor fusion and adapting the system to dynamic environmental conditions. These promising results indicate that the approach reduces setup complexity, lowers costs, and enhances data reliability across diverse environments.

This study develops an automated Slit-Scan Photography (SSP) system addressing four technical limitations in capturing temporal continuity: exposure complexity (16-parameter matrix), reciprocity failure, mechanical instability (±0.01mm precision), and slit contamination. Integrating quaternary exposure controls, servo-driven slit mechanisms, and adaptive ND filters (>14-stop range), the prototype achieves spatiotemporal synthesis of 2D space with 1D time vectors. Grounded in Kolb's experiential learning theory, it materializes Bergson's durée concept while demonstrating commercial potential through patented shutter technology. The system establishes new paradigms for temporal visualization in artistic and scientific imaging, reconciling technical precision with Deleuzian time-image aesthetics.

This talk explores the challenges faced by Netflix Animation Studios in implementing HDR (High Dynamic Range) technology in artist workflows, through an internal animated short film project Sole Mates serving as a case study. We discuss overcoming software and hardware integration constraints, specifically the lack of native HDR support in Linux-based DCC applications when not using dedicated video output, as well as the limited availability of suitable display hardware. Our approach involved developing technical solutions, namely SDR simulation for UI elements and tailored ACES display transforms. These efforts supported creative workflows, enabling artists to work interactively with HDR imagery.

StageCraft, the end-to-end virtual production platform created by Industrial Light & Magic (ILM), is the culmination of decades of experience in the field. Advancements in real-time rendering, combined with motion tracking, have provided an opportunity for a new paradigm: on-set, in-camera visual effects (ICVFX), with a level of spontaneity and flexibility not previously possible in visual effects creation. Filmmakers can now expect near-instanteous responses to their creative direction on set. In order to support StageCraft Operators as they work to achieve a filmmaker's on-demand vision, we developed Oscar: a suite of tablet-based user interfaces which provide intuitive access to the complexities of the underlying StageCraft technology.

We present the Launcher, a software environment configuration tool that has contributed to the success of numerous productions over the course of two decades at Animal Logic, now Netflix Animation Studios. We explore the core features that enable us to manage a large number of configurations across numerous departments and projects while balancing stability and flexibility. We touch on the ability to roll back configurations to any moment in time and how the Launcher helps us with the propagation of VFX platform updates and other new workflows.

At the BYU Center for Animation, we have more students than workstations. Since lab seating is arbitrary, each student needs to be able to work on any machine at any time and each machine must support a wide range of activities, including video game development, animated film production, live class instruction, and homework assignments. For the past three years, we have used VMs in our computer lab disk image to facilitate a hybrid Linux and Windows environment that is flexible, maintainable, and artist-friendly. We present strategies for maximizing performance within the virtual machine, ensuring consistent availability of resources such as file systems and networks between operating systems, securely authenticating, and making the dual-OS workflow artist-friendly. We outline the architecture, challenges faced, and performance benchmarks, demonstrating the potential of VM-enabled, OS-agnostic lab environments for enhancing small studio or educational workflows and maximizing resource utilization.

An R&D initiative within Aardman Animations exploring VP technologies for stop-motion film making. Taking a holistic view of VP from story development through to delivery. Utilising real-time tools, digital twins, and a cross platform XR sandbox. Striving to evolve traditional processes, enhancing creativity, efficiency, and integration across the production pipeline.

In January 2023, FCB New York approached The Mill with a brief for a groundbreaking branded film that would be presented as a TV special. “We want to create an experience for a rookie Formula 1 driver that allows him to virtually race past legends, but in his real car, on a real track. And we want to make it into a TV special.” We knew that this would be one of the most ambitious projects that The Mill has ever undertaken, drawing on every department across several studios around the world.

At the heart of this undertaking was a bespoke Augmented Reality display system for the driver. The final system incorporated race-grade GPS tracking, high-speed data transfer, and a custom display system that was safe enough to mount in the driver's helmet. To bring the race to viewers at home, we leveraged AR camera tracking to match the physical camera positions to the virtual world, creating a seamless viewing experience for fans around the world.

Our groundbreaking groom pipeline for LAIKA’s "Wildwood" revolutionizes stop-motion puppet fabrication through CG-assisted silicone casting. By leveraging the VFX team’s 3D models and digital grooms, we were able to scale to the needs of this epic feature film, while providing anisotropic characteristics and enabling seamless integration between practical and digital elements.

Immersive and interactive media have long sought to engage audiences through sensory-rich experiences. However, designing for public spaces presents unique challenges—such as co-presence (sharing space with others), cognitive overload (excessive sensory input that impairs decision-making), and opaque interfaces (unintuitive or invisible interaction cues), which often hinder engagement. Unlike games or personal screen-based media with clear conventions, experiences in public spaces require intuitive, shared interactions that foster emotional connection between participants to create satisfying immersion.

This paper presents the challenges and innovations behind rigging non-humanoid alien characters with unconventional anatomies—such as limbless, multi-segmented, or fluid forms. Using collaborative workflows, modular rigs, and procedural tools, the team enabled expressive, efficient animation while balancing creative ambition with technical scalability.

From observations of luminous flames, it is evident that soot oxidation behaves as an erosion of the flame. This is due to the underlying physics where the flame interior is oxygen depleted, and oxidation is governed by diffusion and turbulent mixing with the air at the edge of the flame. Motivated by this, we model soot oxidation with a level set equation that combines a physics-based term for naturalistic results with a procedural term providing additional user-control over flame height and thickness. We demonstrate, verify and discuss the capabilities of our method by several examples ranging from small-scale flames to large-scale turbulent fire.

We present a production-proven method for up-resing volumetric fluid simulations, allowing creation of high-resolution results efficiently from fast-to-compute, low resolution sims. Existing approaches often rely on secondary simulation and supplementary turbulence to inject missing details at higher resolutions. At Industrial Light and Magic (ILM), we have devised a simpler approach, one that is frame independent and therefore parallelisable across frames. Our combination of gridless advection, anti-aliased sampling and camera-frustum aligned grids has been used with great success on many recent productions and is a mainstay of our fluids toolkit.

Curves are used to represent assets such as fur, hair, and grass in CG and can be used for rigs that drive vegetation. Applying just noise fields to create procedural wind on curves has been an attractive method. However, this makes the wind appear artificial, lacking in both its dynamic nature that can be art-directed and its realistic interaction with objects. In this talk, techniques are presented to art-direct as well as enhance the realism of procedural curve wind in shots with the addition of collisions, shielding, gusts, and recovery. These provide the artists with significant control over the wind.

Crystals erupt from the ground, surrounding Aang in seconds. To bring this moment to life in Avatar: The Last Airbender, we developed a custom, art-directable crystal growth system built in Houdini that could integrate seamlessly with actors and plates. Designed to support extreme speed, flexible timing, and creative direction, the system enabled us to emphasize the scene's intensity while adhering to the look and logic of real-world crystals.

We present a different approach in modelizing lens characteristics such as vignetting, defocus and chromatic aberrations and how to integrate them in a visual effect pipeline with artistic constraints.

In Pixar’s feature animation Inside Out 2 (2024), the core Emotions led by Joy find themselves navigating through the Stream of Consciousness in an attempt to reach Riley’s Back of the Mind and then retrieve her Sense of Self. Our team was cast to develop the look for the Stream of Consciousness and support its deployment across nearly twenty shots. In particular, it was clear during production that we needed ways to work concurrently with the Animation team, at times even independently, while still able to address notes and changes rapidly. This work discusses our technical solutions for the implementation of the Stream of Consciousness, including in-house procedural tools that facilitated the authoring and stylization of velocity fields interacting with 3D obstacles.

For the interactive beaches featured on Pixar’s Elio, we developed an innovative method which improved upon past approaches. Traditional interactive sand pipelines are made of three parts: a render-optimized surface with limited grain interaction for most of the set, a limited number of interactive particles, and a compositing integration step to bridge a notorious gap between static set sand and interactive sand. Our approach introduces a render-optimized sand asset entirely made of points with reading and writing capability for every point, ready to modify for interaction without look alteration. This drastic workflow simplification provided a much more capable pipeline leading to the efficient authoring of individually tweakable sand grains above half a billion. By leveraging a perceptual model that uses the camera frustums of all frames, we plausibly represented expansive areas which in reality would be composed of many trillions of grains.

We present a suite of techniques from our in-house simulation framework, Loki, addressing the pervasive challenge of collision instabilities in character effects, particularly in cases where nonphysical pinching prevents collision resolution. We introduce a proximity-tolerant mode for contact projection that trades collision residual for stability, a compliant kinematic mechanism for on-demand gap expansion, and contact-aware strain limiting to prevent penetrations while enforcing target edge lengths. Additionally, we showcase our tools for collider management, including hierarchical collision exclusion, one-sided collision handling, and paintable collision thickness maps. These techniques collectively demonstrate a robust and intuitive workflow for combining physically based collisions with challenging production animations.

This work presents a successful real-world transition of a major VFX studio, Rodeo FX, from a multi-software pipeline to a unif ied Solaris and USD-based workflow. Motivated by the growing complexity of episodic and feature film productions, such as Red One and House of the Dragon Season 2, the shift aimed to consolidate disparate departments and eliminate redundant tasks by using Houdini Solaris as the central platform for asset development, Crowd, CFX, FX, lighting, and rendering. The core of this pipeline transformation involved standardizing USD layer stack composition, introducing reusable HDAs and Rodeo Chunks, and automating per-shot workflows using Shotgun event triggers and dispatch graphs. We detail implementation specifics such as the USD Payload Package concept, layer stack auto-generation, and robust automation systems that allowed for efficient iteration, shot synchronization, and cross-departmental collaboration. Finally, we demonstrate production-level results using case studies from House of the Dragon Season 2, including environment builds of King's Landing, Silverwing dragon shading, FX instancing workflows, and automated shot renders. This abstract outlines practical strategies for large studios aiming to migrate to a scalable, artist-friendly, and automation-driven USD pipeline.

We describe a production‑proven workflow that converts deep renders from Dune: Prophecy into compact, photoreal Gaussian‑splat assets for real‑time scouting. A custom Houdini Digital Asset (HDA) assembles COLMAP‑compatible training data from 3 K deep images, then Jawset Postshot trains ∼16 M‑point splats in 3 h on a single RTX 4090. Creatives explored both day and night versions of the kilometer‑scale Imperial Palace set at 60–90 fps, this opened the possibility of digital scouting a full CG environment. The approach delivers the visual fidelity of offline shading while slashing preview latency to seconds and paves the way for using trained splats as final‑pixel backgrounds in compositing.

This paper explores the use of 3D Gaussian Splatting (3DGS) for cultural heritage representation, integrating real-time game engines to create immersive cross-media experiences. Using a historic Hakka mansion in Hong Kong as a case study, the research examines the technical and aesthetic challenges of 3DGS, particularly its limitations in lighting, geometry accuracy, and artifact generation. By embracing these constraints, the project investigates new modes of representing historical spaces beyond traditional digital twins. The study highlights how 3DGS, combined with SLAM-based scanning and game engine workflows, can offer novel approaches to interactive storytelling, site documentation, and artistic reinterpretation in digital heritage preservation and other use cases.

The Metaphysic Neural Performance Toolset introduces a groundbreaking framework for photorealistic, AI-driven human performance synthesis. Leveraging advanced neural architectures, identity-specific training, and latent space manipulation, it delivers unparalleled realism and control for both cinematic post-production and real-time applications. Successfully deployed in major productions such as HERE, Alien: Romulus, and Mad Max: Furiosa, as well as live performances for Drake and Eminem, this toolset redefines AI-generated content in film and entertainment.

This research explores the integration of Generative Artificial Intelligence (GenAI) into immersive, site-specific theater through a collaborative initiative between Moment Factory and Third Rail Projects. The objective was to investigate how GenAI technologies can augment dramaturgical practices by enabling co-creation among authors, performers, and audience members.

Using an iterative research-creation methodology, the project unfolded in two phases: a remote prototyping cycle and an on-site laboratory. Across three use cases: generative scenography, physical space extension, and improvised narration—AI tools were integrated into live performance contexts to dynamically respond to human presence and input. Techniques included the use of real-time generative visuals, silhouette-based transformations, and voice-activated storytelling systems.

The findings highlight both the creative opportunities and technical constraints of employing GenAI in live performance. Critical insights emerged around timing, audience perception, system orchestration, and the need for adaptable, artist-centric AI workflows. This study emphasizes the potential of GenAI to function not merely as a technical instrument, but as a dramaturgical collaborator in the creation of emotionally resonant, participatory experiences.

The main challenge in creating the stylized organic environments of The Wild Robot was to transform our 3D Look Dev Artists into more of a traditional 2D Artist. We wanted to capture the impromptu and organic techniques of 2D art and apply that to a 3D world. A 2D artist is not limited by the technical constraints of a 3D Artist and can express their creativity more freely. In order to recreate the spontaneity of a 2D Artist we needed first to overcome many of the limitations of the 3D medium.

Recent films at DreamWorks Animation, such as The Wild Robot and The Bad Guys 2, have pursued ambitiously illustrative styles, requiring FX with an emphasis on painterly simplification and 2D inspired silhouette and timing, that also maintain a 3D sophistication of scale, detail and rendering. To meet the demanding art direction of these films, while contending with the challenges of 3D Lighting integration and ever-changing cameras in a 3D feature animation pipeline, we developed a suite of tools to let artists creatively mix simulation with drawing all from within a modern node-based FX software package. The collection of tools, called Doodle, enable artists to use traditional painting and sculpting techniques in 3D space, procedurally manipulate and simulate the resulting geometry, and compose these hand-drawn elements into reusable 3D sprites for aggregate effects.

The environment in The Wild Robot is rich in variety, dense in layout, painterly in look, and a character in itself alive with motion. This talk describes the different techniques, tools, and pipeline used to breathe life into this wild environment, along with the challenges faced while dealing with the environment’s complexity.

The production design for Pixar’s Elio (2025) called for complex and dynamic elements throughout the "Communiverse" environments of the film. In this talk we showcase several of these effects and discuss the differences in implementation to acheive specific visual goals, workflow efficencies and render optimizations. Many of these approaches blurred the lines between traditional disciplines and departments such as modeling, shading, lighting, dressing and animation. We leveraged features in Houdini, and USD, as well as custom shader solutions in Pixar’s proprietary look development environment Flow to achieve the dynamic features in the film.

The aliens in Pixar’s Elio commanded a unique look in space. We crafted each species individually with respect to its own characteristic features, carefully choosing colors and material properties to create over 18 species. Their designs took inspiration from various flora and fauna including micro-organisms, worm-like, bird-like, and sea creatures. They needed to be appealing, organic, and tactile but without feeling "earth bound”. To implement these designs and help them feel unique and alien, our team explored combining illumination models and animated shading techniques in ways not commonly found in earth species and materials. This process created a collaborative and interactive approach to our look development to celebrate diversity and inclusivity in Elio’s fantastic space aliens.

In Pixar’s Elio (2025), Elio visits the Communiverse, a protopian space station populated by various species from throughout the universe. To convey a vivacious city inhabited by a diverse population, a small team of environment artists amplified the traditional pipeline with procedural techniques used in unique ways to develop a vast quantity of various biomes of alien terrains and architectures that pulsed with their own internal energy.

Lighting is a critical part of our animated films. We developed a new Lighting workflow in Houdini Solaris for the feature production, Walt Disney Animation Studios’ "Moana 2". Our goal was to empower artists with greater control while reducing barriers to creative expression. How do we enable new workflows with greater control while lowering the barrier for our artists to bring their visions to life? What experiences do we mirror in the new tooling landscape and what do you leave behind? This talk explores the lessons learned and some best practices for building and adopting new tools within a legacy system while creating a seamless experience for our artists. We will also look at a few example toolsets we developed to help our Lighting artists transition to the new workflow.

Amongst a variety of alien species in Pixar’s Elio, one of the more unusual characters in the film is the Universal Users Manual (or UUM for short) - a sentient alien book harboring all the answers to the universe.

The UUM is a “character” consisting of a stack of pages with the ability to zoom along distinctly shaped paths and is rigged and animated in our traditional pipeline within Presto with the additional hookup of Houdini Engine (HE) to create the distinct look of the character with traditional FX layered on top. By hooking up a base shape to Houdini Engine we have the ability to combine native ease of manipulating a character in presto with the power and sophistication of our FX pipeline.

We present a novel character rigging solution developed for OOOOO, a liquid supercomputer in Pixar’s Elio.

OOOOO’s design and desired movement necessitated reimagining our conventional way of articulating characters and she became Pixar’s first mesh-free character rig. We developed a system that allowed our animators full fidelity control over what is essentially a rigged shader while ensuring downstream renderability. [Luo et al. 2025] The system’s architecture supports a hierarchical arrangement of implicit surface primitives and operators, allowing for complex transformations while preserving normal animation paradigms and offers unprecedented flexibility in character animation.

How do you shade a liquid supercomputer created as an implicit surface? Elio’s OOOOO is Pixar’s first character made and rigged entirely as a controllable series of metaball-like signed distance functions, rendered for interaction with a GLSL shader [Lykkegaard et al. 2025] in our animation software, Presto. Although this novel approach facilitates incredible animation, it does not provide a stable mesh for shading, rendering, and deformation purposes. OOOOO is made of a body and blobs, all capable of separating and merging at any time, filled with “nests” of moving circuits that are concentrated at the core. This talk addresses the accompanying set of challenges these factors created for look development past the model/rig stage, resulting in a per-frame process.

Figure 1:Elio’s OOOOO has an implicit surface rig that can take many shapes ©Pixar

Cloning Clay, a space amoeba-like organic matter generating a variety of laughs, thrills, and clones throughout Pixar’s Elio (2025), required a suite of technical and creative FX techniques to land each story beat with a satisfying performance. In regular collaboration with several departments, this method delivered a range of effects, including dynamic hero clay FX, secondary rippling, and full-character transformations. A lightweight Houdini-based workflow was developed to ensure these techniques scaled efficiently across multiple sequences while minimizing per-shot overhead. Most shots were delivered via this base setup, but full-body transformations required significant rig customizations to handle increased complexity.

We propose AniDepth, a novel anime in-betweening method using a video diffusion model. Even if the model is fine-tuned on an anime dataset, it still suffers from the domain gap between anime and the natural image domain, due to the strong priors of the base model. Therefore, we convert anime illustration into depth map, which is a modality filling the gap between anime and realistic domains to fully exploit the model’s prior knowledge. In addition, by using line-arts as guide during the in-betweening, we enhance the fidelity of the generated line-arts details. Our approach first interpolates converted depth maps, warps line-arts based on the depth maps, then interpolates colored images with using the line-arts as the conditions, preserving complex line-art and flat coloring even under large movements. Experiments show that our method outperforms the existing diffusion method in quantitative evaluation, improves temporal smoothness, and reduces line-art distortion. Moreover, because it requires no extra training, it can be easily integrated into current production pipelines at low operational cost.

This paper describes the techniques we use to build the complex light rigs in the Sony Pictures Imageworks lighting pipeline. Typically we receive the panoramic HDRI from the set and need to make a light rig from it. Building a light rig has several steps: extracting area lights from HDRI, placing them in a 3D scene, aligning lights with Lidar from set. We describe how this process can be sped up. For example, the positioning of extracted lights is automated using PatchFusion [Li et al. 2023], an off-the-shelf machine learning model for high resolution monocular depth estimation. PatchFusion computes accurate metric depth directly from the HDRI. The depth map is used as a distance from light to camera to place the area lights in 3D scene. This removes the need for manual distance measurements or guesswork. Our approach significantly reduces manual labor. The time required to build the light rig goes from hours to about a minute.

This talk explores how modern foundation models, particularly vision-language models (VLMs), can intelligently interface with and augment 3D graphical tools like Blender. Through three major contributions — BlenderAlchemy, BlenderGym, and FirePlace — we investigate the potential of AI-driven automation in 3D content creation while ensuring human control and editability. We show that even though VLMs of today have many weaknesses, inference-compute scaling and complementary tool-use are powerful methods that can improve their performance on graphical editing tasks in a training-free manner. Through the works I’ll share 4 different ways of scaling inference compute to improve VLMs’ capabilities on operating human graphical tools to accomplish graphical edits.

We describe a novel method that improves Digital Domain’s hybrid video-driven animation transfer technique for facial motion capture [Serra et al. 2022]. In this work, we automate the animation correction pass, otherwise done by artists, and accelerate the production cycle while minimizing subjectivity associated with matching the actor’s facial expressions to the CG character. We leverage our video-driven animation transfer model that produces images of the CG character matching the actor’s performance, by using those images as targets in a differentiable renderer optimization loop. Thus improving the model’s initially predicted geometry. Furthermore, lighting parameters are removed from the optimization by training light-invariant models with a simple augmentation strategy. The corrected animations can be used directly in shots or to fine-tune the base model, as done in the earlier approach. Validation tests confirmed our method’s efficacy, and it is now being integrated into Digital Domain’s facial motion capture workflow.

Virtual crowds are prevalent in entertainment media, including movies, games, and educational or training experiences. In this work, we analyze algorithms for simulating such crowds. Specifically, we demonstrate that widely popular state-of-the-art algorithms fail in several basic benchmark cases. With the goal of designing more robust algorithms, we discuss potential solutions, which can be easily integrated into the specified simulation crowds.

Crowds scenes, in which multiple characters come to life to achieve a cohesive performance, present a unique set of animation challenges. While the techniques, workflows, organization and execution have all evolved for more than 100 years at Walt Disney Animation Studios, the goals, the underlying principles, and the fundamentals have remained the same. We present a distillation of the essence of the art of crowds animation with examples from a wide range of Disney’s animated films.

Collisions are a key problem in generating complex crowds animation. The mudskippers in Walt Disney Animation Studios’ "Moana 2" presented a particularly challenging scenario where throngs of floppy amphibious fish pack tightly together to form a towering pile. Our solution introduces an additional simulation step to deform the skinned character meshes using Houdini’s Vellum solver to resolve body-to-body contact. This departure from our standard deferred mesh generation workflow provided an effective solution to this problem, and opens up new avenues for achieving character deformations beyond the capabilities of character rigging.

In Walt Disney Animation Studios’ "Moana 2", our main character, Moana, embarks on a journey which takes her farther from home than she has ever been before. She undergoes a mental odyssey of dealing with what it means to be so distant from those she loves. Moana’s voyage takes her through magical and mystical worlds that greatly differ in visual style. This visual design and the camera language used across the songs of "Moana 2" support the progression of Moana’s character and the cinematography celebrates the distinct look and feel of these unique worlds.

We present a procedural, modular, and asset-centric approach to character effects and simulation (CFX) workflow particularly regarding costumes and hair/fur. This differs from many other approaches by focusing on the simulation setups at individual asset level, and then using costume collection configuration, procedurally constructing the simulation scene by merging the assets and having them solved together. Shot specific modifications and overrides such as custom cloth poses and parameter tweaks can be applied in a hierarchical and procedural manner so that when there are upstream changes, the subsequent job iterations will pick up the changes without requiring manual rework.

Live physical whole-puppet performances can be used to drive digital animation characters and creatures via puppix, a new capture system. The benefits of having a live puppet character in the room with actors, directors and other characters are demonstrated and discussed, as well as practical processes of capturing non-human physicalities.

We practically demonstrate the experience of having a live puppet performance in the room via audience interaction with a professionally performed theatrical puppet (Mr D Wolf), which continues at points during the presentation.

Physical puppets allow directors and actors to work with non-human characters with the same flexibility, freedom and immediacy as human actors. Capturing these performances means non-human digital characters can work alongside and be directed like physical actors.

Unlike keyframe animation, capture of in-the-moment live performance allows real-world weight, physicality and movement transfer to digital twins. This also disrupts the limitations of human-based motion capture systems and the bulk of learning model training data sets, whose movements are originally from human physicality.

The origins of motion capture as a whole come from the technology of puppetry and animatronics. Performance armatures and rigs like Dinosaur Input Device, Sil and Hensons’ Waldo operate as control systems for digital performances, with director focus on digital output screens.

puppix, a whole-puppet capture system, keeps the performance focus on the character in the room, not on the screen.

We show examples of pre-recorded puppix outputs alongside the original performance footage.

We detail practicalities: of how to build a puppet pair for motion capture, successful live performance methods and processes, data processing considerations, practical capture considerations and extra notes for the VFX supervisor.

At present, reference puppets are being used on-set as placeholders for CG characters, providing lighting, position, interaction reference, whilst actors interact with the reference puppeteers’ performances.

puppix allows the full reference puppet performance to be motion captured. Secondary movements and whole body physicality match the digital characters and transfer to the digital character for free. Director and performers focus in the room, whilst creating digital animated performances. This is a tool like human-based motion capture, but for non-human characters and creatures.

Creating a believable digital character that has never existed requires a large team of dedicated people. We often have real-world source material to reference, but in the case of Malgosha, there was nothing but a concept and a charmingly low-resolution game world. That's when we rely on Wētā FX's recipe of collaboration working towards a common goal of making audiences believe that something unbelievable is real.

Malgosha, the main villainess, evolved from a simple concept into a fully realized digital character. We will showcase the intricate and collaborative process that brought her to life, highlighting Wētā FX's unique approach and deep history in character creation.

Achieving lifelike character deformations in high-end visual effects is both an artistic and technical challenge. This paper presents BodyOpt [Sprenger et al. 2023] 2.0, WetaFX’s advanced character deformation pipeline, focusing on generalization beyond bipeds, improved ease of use, and higher-quality reconstructions and dynamics. We highlight technical advancements, including a more efficient model transfer process, and discuss how BodyOpt 2.0 enables scalable production workflows across small and large projects.