This year at SIGGRAPH 2025, Sebastian Herholz from Intel organized a followup to 2019’s Path Guiding in Production Course [Vorba et al. 2019]. This year’s edition of the course includes presentations by Sebastian on Intel’s Open Path Guiding library and on general advice for integrating path guiding techniques into a unidirectional path tracing renderer, a presentation by Martin Šik on how Chaos’s Corona Renderer uses advanced photon guiding techniques in their caustics solver, and a presentation by Lea Reichardt and Marco Manzi on the work Disney Animation and DisneyResearch|Studios have put into Hyperion’s second-generation path guiding system for surfaces and volumes. I strongly encourage checking out the whole course, but wanted to highlight Lea and Marco’s presentation in particular; they put a ton of care and effort into what I think is a really cool and unique look into what it takes to bridge cutting edge research into a production rendering environment. The course notes were written by the four presenters above, in addition to Brian Green and myself from the Hyperion development team.

The course will be presented on Tuesday August 12th, startng at 3:45 PM.

Here is the abstract:

We present our approach to implementing a second-generation path guiding system in Disney’s Hyperion Renderer, which draws upon many lessons learned from our earlier first-generation path guiding system. We start by focusing on the technical challenges associated with implementing path guiding in a wavefront style path tracer and present our novel solutions to these challenges. We will then present some powerful visualization and debugging tools that we developed along the way to both help us validate our implementation’s correctness and help us gain deeper insight into how path guiding performs in a complex production setting. Deploying path guiding in a complex production setting raises various interesting challenges that are not present in purely academic settings; we will explore what we learned from solving many of these challenges. Finally, we will look at some concrete production test results and discuss how these results inform our large scale deployment of path guiding in production. By providing a comprehensive review of what it took for us to achieve this deployment on a large scale in our production environment, we hope that we can provide useful lessons and inspiration for anyone else looking to similarly deploy path guiding in production, and also provide motivation for interesting future research directions.

The paper and related materials can be found at:

• Project Page (Author’s Version and Demo Video)
• Official Course Website
• Official Print Version (ACM Library)

All of the technical details are in the paper and presentation (and with 80 pages of course notes, of which 36 pages is the Disney Animation / DisneyResearch|Studios chapter, there are technical details for days!), so this blog post is just some personal thoughts on this project.

As mentioned in the abstract, what we’re presenting in this course is our second-generation path guiding system. Disney Animation and DisneyResearch|Studios have a long history of working on path guiding; one of landmark papers in modern path guiding was Practical Path Guiding (PPG) [Müller et al. 2017], which came out of DisneyResearch|Studios, and Hyperion was one of the first production renderers to implement PPG [Müller 2019]. We’ve used path guiding on a limited number of shots on most movies starting with Frozen 2, but as the course notes go into more detail on, for a variety of reasons our first generation path guiding system never gained widespread adoption. Several years ago, based on a research proposal drafted by Wei-Feng Wayne Huang while he was still at Disney Animation, we kicked off of a large scale project to further improve path guiding and bring it to a point where we could get widespread adoption and provide significant benefits to production. This project is a collaboration between DisneyResearch|Studios, Disney Animation, Pixar, ILM, and Sebastian Herholz from Intel; the second-generation path guiding system in Hyperion is a product of this project.

Personally, this project is one of my all-time favorite projects that I’ve gotten to be a part of, and I think this project really highlights what an incredible research organization DisneyResearch|Studios is. Getting to collaborate with rendering colleagues from across multiple labs and studios is always fun and interesting, and I think for projects like this, the huge amount of production and engineering expertise that the various Disney studios bring combined with the world-class research talent at DisneyResearch|Studios and ETH Zürich (which DisneyResearch|Studios is academically partnered with) gives unique perspectives and capabilities for tackling difficult problems. On top of the production experience from three cutting edge studios, DisneyResearch|Studios also has deep access to both source code and the engineering teams for not one but two extremely mature production rendering systems, Hyperion [Burley et al. 2018] and RenderMan [Christensen et al. 2018]; I don’t think there’s anything else quite like this setup in our industry, and I think it’s insanely cool that we get to work together like this!

One of the largest focus points for our second-generation path guiding efforts was to find a way to guide jointly for both surfaces and volumes; our first-generation PPG based system only supported surfaces. Over the past decade our artists have made heavier and heavier use of volumetrics in each successive project, to the point where now almost every shot in our movies contains some form of volumetrics, ranging from subtly atmospherics all the way to enormously complex setups like the storm battle at the end of Moana 2. We already knew from past experience that extending PPG to volumes wasn’t as easy as it might look, and a second-generation path guiding system would likely need to be a significant departure from PPG. Towards the start of this project we learned that Sebastian Herholz at Intel was working in a similar direction and had incorporated a wide swath of recent path guiding research [Müller et al. 2017, Herholz et al. 2019, Ruppert et al. 2020, Xu et al. 2024] into Intel’s open source OpenPGL library; at this point the project was expanded to include a collaboration with Sebastian. This collaboration has been extraordinarily fruitful, with work from the Disney side of things helping inform development for OpenPGL and expertise from Sebastian helping us build on top of OpenPGL.

An interesting aspect of our next-generation path guiding project is that this project has been both an academic research project and a production engineering project; over the past several years, this project has spawned a series of cool research papers, but has also included a huge effort to get all of this research implemented in Hyperion and RenderMan and into artists’ hands to actually make movies with, which means solving tons of practical production problems that sit outside of the usual research focus. On the research side, so far the project has spawned three papers. Dodik et al. [2022] improved upon PPG by using spatio-directional mixture models to improve guiding’s ability to learn and product sample arbitrarily oriented BSDFs. Xu et al. [2024] introduces a way to guide volume scattering probaiblity (which is indirectly related to distance sampling) in volume rendering, which historically has been a major missing piece in path guiding in volumes. The most recent paper, Rath et al. [2025], looks at how to incorporate GPU based neural forms of path guiding into existing CPU based renderers. Each of these research papers tackles a major challenge we’ve found while working towards making path guiding practical in production.

To bridge between the research work and building a practical production system, we’ve put a lot of work into solving both more architectural technical challenges, and more artist facing user experience challenges. One of the largest architectural technical challenges has been fitting path guiding, which learns from full path histories, into Hyperion’s wavefront rendering architecture [Eisenacher et al. 2013], in which path histories are not kept beyond the current bounce for each path (RenderMan XPU [Christensen et al. 2025] is also a wavefront system, so the challenges there are similar). Artist facing user experience challenges involve the reality that production renderers include many features that break physics to allow for better artistic control and more predictable results, which are difficult to account for when developing path guiding techniques in a purely academic renderer. Solving these engineering and user experience challenges in order to build a practical production system is the focus of our part of the course this year. What we’re presenting in the course is really a snapshot of where we were at the beginning of the year; the material in the course represents enormous progress towards a robust practical system, but this project is very much still in progress and we’ve made additional advancements since we finished writing the course materials; hopefully we can present even more next year!

My favorite part of working on these projects is always getting to work with and learn from really cool people. On the DisneyResearch|Studios side, this project has been led by Marco Manzi and Marios Papas, with significant contributions by Alexander Rath, Tiziano Portenier, and engineering support from Lento Manickathan. On the Disney Animation side, Lea Reichardt, Brian Green, and I have been working closely with Marco to build out the production system. Of course we have a long list of artists and TDs to thank on the production side for supporting and trying out this project; a full list of acknowledgements can be found on my project page and in the course notes. Pushing path guiding forward in research and production together has been the type of project where everyone on the project learns a lot from each other, the studios learn a lot and gain incredibly powerful and useful new tools, and the wider research community benefits as a whole. Investing in these types of large scale, longer term research projects is always a daunting task, and the fact that our studio leadership has given so much support this project and given us the time and resources to really make it a big success is just another testament towards the commitment Disney as a whole has towards making the best movies we can possibly make!

References

Brent Burley, David Adler, Matt Jen-Yuan Chiang, Hank Driskill, Ralf Habel, Patrick Kelly, Peter Kutz, Yining Karl Li, and Daniel Teece. 2018. The Design and Evolution of Disney’s Hyperion Renderer. ACM Transactions on Graphics. 37, 3 (2018), Article 33.

Per H. Christensen, Julian Fong, Jonathan Shade, Wayne L Wooten, Brenden Schubert, Andrew Kensler, Stephen Friedman, Charlie Kilpatrick, Cliff Ramshaw, Marc Bannister, Brenton Rayner, Jonathan Brouillat, and Max Liani. 2018. RenderMan: An Advanced Path Tracing Architecture for Movie Rendering. ACM Transactions on Graphics. 37, 3 (2018), Article 30.

Per H. Christensen, Julian Fong, Charlie Kilpatrick, Francisco Gonzalez Garcia, Srinath Ravichandran, Akshay Shah, Ethan Jaszewski, Stephen Friedman, James Burgess, Trina M. Roy, Tom Nettleship, Meghana Seshadri, and Susan Salituro, 2025. RenderMan XPU: A Hybrid CPU+GPU Renderer for Interactive and Final-Frame Rendering. Computer Graphics Forum (Proc. of High Performance Graphics) 44, 8 (Jun. 2025), Article e70218.

Ana Dodik, Marios Papas, Cengiz Öztireli, and Thomas Müller. 2022. Path Guiding Using Spatio-Directional Mixture Models. Computer Graphics Forum (Proc. of Eurographics) 41, 1 (Feb. 2022), 172-189.

Christian Eisenacher, Gregory Nichols, Andrew Selle, and Brent Burley. 2013. Sorted Deferred Shading for Production Path Tracing. Computer Graphics Forum. 32, 4 (2013), 125-132.

Sebastian Herholz, Yangyang Zhao, Oskar Elek, Derek Nowrouzezahrai, Hendrik P. A. Lensch, and Jaroslav Křivánek. 2019. Volume Path Guiding Based on Zero-Variance Random Walk Theory. ACM Transactions on Graphics (Proc. of SIGGRAPH) 38, 3 (Jun 2019), Article 25.

Thomas Müller, Markus Gross, and Jan Novák. 2017. Practical Path Guiding for Efficient Light-Transport Simulation. Computer Graphics Forum (Proc. of Eurographics Symposium on Rendering) 36, 4 (Jun. 2017), 91-100.

Thomas Müller. 2019. Practical Path Guiding in Production. In ACM SIGGRAPH 2019 Course Notes: Path Guiding in Production. 37-50.

Alexander Rath, Marco Manzi, Farnood Salehi, Sebastian Weiss, Tiziano Portenier, Saeed Hadadan, and Marios Papas. 2025. Neural Resampling with Optimized Candidate Allocation. In Proc. of Eurographics Symposium on Rendering (EGSR 2025). Article 20251181.

Lukas Ruppert, Sebastian Herholz, and Hendrik P. A. Lensch. 2020. Robust Fitting of Parallax-Aware Mixtures for Path Guiding. ACM Transactions on Graphics (Proc. of SIGGRAPH) 39, 4 (Aug 2020), Article 147.

Jiří Vorba, Johannes Hanika, Sebastian Herholz, Thomas Müller, Jaroslav Křivánek, and Alexander Keller. 2019. Path Guiding in Production. In ACM SIGGRAPH 2019 Courses. Article 18.

Kehan Xu, Sebastian Herholz, Marco Manzi, Marios Papas, and Markus Gross. 2024. Volume Scattering Probability Guiding. ACM Transactions on Graphics (Proc. of SIGGRAPH Asia) 43, 6 (Nov. 2024), Article 184.

This year at SIGGRAPH 2025, Mark Lee, Nathan Zeichner, and I have a talk about a GPU texture streaming system we’ve been working on for Disney Animation’s in-house real-time GPU ray tracing previsualization renderer. Of course, GPU texture streaming systems are not exactly something novel; pretty much every game engine and every GPU-based production renderer out there has one. However, because Disney Animation’s texturing workflow is 100% based on Ptex, our texture streaming system has to be built to support Ptex really well, and this imposes some interesting design requirements and constraints on the problem. We thought that these design choices would make for an interesting talk!

Nathan will be presenting the talk at SIGGRAPH 2025 in Vancouver as part of the “Real-Time and Mobile Techniques” session on Sunday August 10th, starting at 9am.

Here is the paper abstract:

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.

The paper and related materials can be found at:

• Project Page (Author’s Version and Demo Video)
• Official Print Version (ACM Library)

As usual, all of the technical details are in the paper and presentation, so this blog post is just my personal notes on this project.

We’ve been working on this project for a pretty long while, and what we’re presenting in this talk is actually the second generation our team has built of a GPU texture streaming system with Ptex support. The earlier first prototype of our GPU Ptex system was largely written by Joe Schutte, who we are very indebted to for paving the way and proving out various ideas, such as the use of cuckoo hashing [Erlingsson 2006] for storing keys. We learned a ton of lessons from that first prototype, which informed the modern incarnation of the system. The core of the modern system was primarily written by Mark, with a lot of additional work from Nathan to generalize the system to support both our CUDA/Optix based GPU ray tracing previsualization renderer, and our in-house fork of Hydra’s Storm rasterizer. My role on the project was pretty small; I essentially was just a consultant contributing to some ideas and brainstorming, so I’m very grateful to Mark and Nathan for having allowed me to contribute to this talk!

One of the biggest lessons I’ve learned during my professional career has been the value of building systems twice- Chapter 11 of the famous Mythical Man Month book by Frederick Brooks is all about the value of building a first version to throw away, because much of what is required to build an actually good solution can only be learned through the process of attempting to build a solution in the first place. A lot of the design choices that went into the system described in our talk draws from both the earlier prototype that was built, and also upon past experience building texture streaming systems in other renderers. For example, one big lesson that Mark and I both learned independently is that texture filtering is extremely hard (and it’s famously even harder in Ptex due to the need to filter across faces with potentially very different resolutions), and in a stochastic ray tracing renderer, a better solution is often to just point sample and lineraly interpolate between the two closest MIP levels. Mark learned this on Moonray [Lee et al. 2017], and I’ve written about learning this on the blog before. I think this project is a great example of both learning from previous attempts at the same general problem domain, but also avoiding the second-system effect; what we have today is really fast, really robust, and given how hard texture streaming generally is as a problem domain, I think Mark and Nathan did an impressive job in keeping the actual implementation compact, elegant, and easy to work with.

Of course we need to acknowledge that there have in fact been previous attempts at implementing Ptex on the GPU, with varying degrees of success. McDonald & Burley [2011] was the first demonstrated implementation of Ptex on the GPU, but required a preprocessing step and had to deal with various complications imposed by using OpenGL/DirectX hardware texturing; this early implementation also didn’t support texture streaming. Our implementation is built primarily in CUDA and bypasses all of the traditional graphics API stuff for texturing; we deal with the per-face textures at the raw memory block level, which allows us to have zero preprocessing steps and have robust fast streaming from the CPU to the GPU. Kim et al. [2011]’s solution was to pack all of the individual per-face textures into a single giant atlased texture; back when I worked on Pixar’s Optix-based preview path tracer (called RTP) this essentially was the solution that we used. However, this solution faces major problems with MIP mapping, since faces that are next to each other in the atlas but non-adjacent in the mesh topology can bleed into each other while filtering to generate each level in the MIP chain for the single giant atlas. By streaming the original per-face textures to the GPU and using exactly the same data as what’s in the CPU Ptex implementation, we avoid all of the issues with atlasing.

An interesting thing that I think our system demonstrates is that some of the preexisting assumptions about Ptex that float around in the wider industry aren’t necessarily true. For some time now there’s been an assumption that Ptex cannot be fast for incoherent access; while it is true that Hyperion gains performance advantages from coherent shading [Eisenacher et al. 2013] and therefore coherent Ptex reads, I don’t think this is really a property of Ptex itself (as hinted at by PBRT’s integration of Ptex). One notable thing about our interactive GPU path tracing use case is that the Ptex access pattern is totally incoherent for secondary bounces- we use depth-first integrators in our previz path tracer. The demo video we included with the talk doesn’t really show this since in the demo video we just show a headlight-shaded view for the sake of clearly illustrating the texture streaming behavior, but in actual production usage our texture streaming system serves multi-bounce depth-first path tracing at interactive rates without a problem.

A final note on the demo video- unfortunately I had to capture the demo video over remote desktop at the time, so there are a few frame hitches and stalls in the video. Those hitches and stalls come entirely from recording over remote desktop, not from the texture streaming system; in Nathan’s presentation, we have some better demo videos that were recorded via direct video capture off of HDMI, and in those videos there are zero frame drops or stalls even when we force-evict the entire contents of the on-GPU texture cache.

I want to thank both the Hyperion development and Interactive Visualization development teams at Disney Animation for supporting this project, and of course we thank Brent Burley and Daniel Teece for their feedback and assistance with various Ptex topics. Finally, thanks again to Mark and Nathan for being such great collaborators. I’ve had the pleasure of working closely with Mark on a number of super cool projects over the years and I’ve learned vast amounts from him. Nathan and I go back a very long way; we first met in school and we’ve been friends for around 15 years now, but this was the first time we’ve actually gotten to do a talk together, which was great fun!

That’s all of my personal notes for this talk. If this is interesting to you, please go check out the paper and catch the presentation either live at the conference or recorded afterwards!

References

Frederick P. Brooks, Jr. 1975. The Mythical Man-Month: Essays on Software Engineering, 1st ed. Addison-Wesley.

Brent Burley and Dylan Lacewell. 2008. Ptex: Per-face Texture Mapping for Production Rendering. Computer Graphics Forum (Proc. of Eurographics Symposium on Rendering) 27, 4 (Jun. 2008), 1155-1164.

Brent Burley, David Adler, Matt Jen-Yuan Chiang, Hank Driskill, Ralf Habel, Patrick Kelly, Peter Kutz, Yining Karl Li, and Daniel Teece. 2018. The Design and Evolution of Disney’s Hyperion Renderer. ACM Transactions on Graphics. 37, 3 (2018), Article 33.

Christian Eisenacher, Gregory Nichols, Andrew Selle, and Brent Burley. 2013. Sorted Deferred Shading for Production Path Tracing. Computer Graphics Forum (Proc. of Eurographics Symposium on Rendering) 32, 4 (Jul. 2013), 125-132.

Ulfar Erlingsson, Mark Manasse, and Frank Mcsherry. 2006. A Cool and Practical Alternative to Traditional Hash Tables. Microsoft Research Tech Report.

Sujeong Kim, Karl Hillesland, and Justin Hensley. 2011. A Space-efficient and Hardware-friendly Implementation of Ptex. In ACM SIGGRAPH Asia 2011 Sketches. Article 31.

Mark Lee, Brian Green, Feng Xie, and Eric Tabellion. 2017. Vectorized Production Path Tracing. In Proc. of High Performance Graphics (HPG 2017). Article 10.

John McDonald and Brent Burley. 2011. Per-Face Texture Mapping for Real-time Rendering. In ACM SIGGRAPH 2011 Talks. Article 10.

This fall marked the release of Moana 2, Walt Disney Animation’s 63rd animated feature and the 10th feature film rendered entirely using Disney’s Hyperion Renderer. Moana 2 brings us back to the beautiful world of Moana, but this time on a larger adventure with a larger canoe, a crew to join our heroine, bigger songs, and greater stakes. The first Moana was at the time of its release one of the most beautiful animated films ever made, and Moana 2 lives up to that visual legacy with frames that match or often surpass what we did in the original movie. I got to join Moana 2 about two years ago and this film proved to be an incredibly interesting project!

While we’ve now used Disney’s Hyperion Renderer to make several sequels to previous Disney Animation films, Moana 2 is the first time we’ve used Hyperion to make a sequel to a previous film that also used Hyperion. From a technical perspective, the time between the first and second Moana films is filled with almost a decade of continual advancement in our rendering technology and in our wider production pipeline. At the time that we made the first Moana, Hyperion was only a few years old and we spent a lot of time on the first Moana fleshing out various still-underdeveloped features and systems in the renderer. Going into the second Moana, Hyperion is now an extremely mature, extremely feature rich, battle-tested production renderer with which we can make essentially anything we can imagine. Almost every single feature and system in Hyperion today has seen enormous advancement and improvement over what we had on the first Moana; many of these advancements were in fact driven by hard lessons that we learned on the first Moana! Compared with the first Moana, here’s a short, very incomplete laundry list of improvements made over the past decade that we were able to leverage on Moana 2:

• Moana 2 uses a completely new water rendering system that represents an enormous leap in both render-time efficiency and easier artist workflows compared with what we used on the first Moana; more on this later in this post.
• After the first Moana, we completely rewrote Hyperion’s previous volume rendering subsystem [Habel 2017] from scratch; the modern system is a state-of-the-art delta-tracking system that required us to make foundational research advancements in order to implement [Kutz et al. 2017, Huang et al. 2021].
• Our traversal system was completely rewritten to better handle thread scalability and to incorporate a form of rebraiding to efficiently handle gigantic world-spanning geometry; this was inspired directly by problems we had rendering the huge ocean surfaces and huge islands in the first Moana [Burley et al. 2018].
• On the original Moana, ray self-intersection with things like Maui’s feathers presented a major challenge; Moana 2 is the first film using our latest ray self-intersection prevention system that notably does away with any form of ray bias values.
• We introduced a limited form of photon mapping on the first Moana that only worked between the sun and water surfaces [Burley et al. 2018].; Moana 2 uses an evolved version of our photon mapper that supports all of our light types, many or our standard lighting features, and even has advanced capabilities like a form of spectral dispersion.
• We’ve made a number of advancements [Burley et al. 2017, Chiang et al. 2016, Chiang at al. 2019, Zeltner et al. 2022] to various elements of the Disney BSDF shading model.
• Subsurface scattering on the first Moana was done using normalized diffusion; since then we’ve moved all subsurface scattering to use a state-of-the-art brute force path tracing approach [Chiang et al. 2016].
• Eyes on the first Moana used our old ad-hoc eye shader; eyes on Moana 2 use our modern physically plausible eye shader that includes state-of-the-art iris caustics calculated using manifold next event estimation [Chiang & Burley 2018].
• The emissive mesh importance sampling system that we implemented on the first Moana and our overall many-lights sampling system has seen many efficiency improvements [Li et al. 2024].
• Hyperion has gained many more powerful features granting artists an enormous degree of artistic control both in the renderer and post-render in compositing [Burley 2019, Burley et al. 2024].
• Since the first Moana, Hyperion’s subdivision/tessellation system has gained an advanced fractured mesh system that makes many of the huge-scale effects in the first Moana movie much easier for us to create today [Burley & Rodriguez 2022].
• We’ve introduced path guiding into Hyperion to handle particularly difficult light transport cases [Müller et al. 2017, Müller 2019].
• The original Moana used our somewhat ad-hoc first-generation denoiser, while Moana 2 uses our best-in-industry, Academy Award winning¹ second-generation deep learning denoiser jointly developed by Disney Research Studios, Disney Animation, Pixar, and ILM [Vogels et al. 2018, Dahlberg et al. 2019].
• Even Hyperion’s internal architecture has changed enormously; Hyperion originally was famous for being a batched wavefront renderer, but this has evolved significantly since then and continues to evolve.

There are many many more changes to Hyperion that there simply isn’t room to list here. To give you some sense of how far Hyperion has evolved between Moana and Moana 2: the Hyperion used on Moana was internally versioned as Hyperion 3.x; the Hyperion used on Moana 2 is internally versioned as Hyperion 16.x, with each version number in between representing major changes. In addition to improvements in Hyperion, our rendering team has also been working for the past few years on a next-generation interactive lighting system that extensively leverages hardware GPU ray tracing; Moana 2 saw the widest deployment yet of this system; I can’t say much more on this topic yet but hopefully we’ll have more to share soon.

Outside of the rendering group, literally everything else about our entire studio production pipeline has changed as well; the first Moana was made mostly on proprietary internal data formats, while Moana 2 was made using the latest iteration of our cutting-edge modern USD pipeline [Miller et al. 2022, Vo et al. 2023, Li et al. 2024]. The modern USD pipeline has granted our pipeline many amazing new capabilities and far more flexibility, to the point where it became possible to move our entire lighting workflow to a new DCC for Moana 2 without needing to blow up the entire pipeline. Our next-generation interactive lighting system is similarly made possible by our modern USD pipeline. I’m sure we’ll have much more about this at SIGGRAPH!

While I get to work on every one of our feature films and get to do fun and interesting things every time, Moana 2 is the most direct and deep I’ve worked on one of our films probably since the original Moana. There are two specific projects I worked on for Moana 2 that I am particularly proud of: a completely new water rendering system that is part of Moana 2’s overall new water FX workflow, and the volume rendering work that was done for the storm battle in the movie’s third act.

On the original Moana, we had to develop a lot of custom water simulation and rendering technology because commercial tools at the time couldn’t quite handle what the movie required. On the simulation side, the original Moana required Disney Animation to invent new techniques such as the APIC (affine particle-in-cell) fluid simulation model [Jiang et al. 2015] and the FAB (fluxed animated boundary) method for integrating procedural and simulated water dynamics [Stomakhin and Selle 2017]. Disney Animation’s general philosophy towards R&D is that we will develop and invent new methods when needed, but will then aim to publish our work with the goal of allowing anything useful we invent to find its way into the wider graphics field and industry; a great outcome is when our publications are adopted by the commercial tools and packages that we build on top of. APIC and FAB were both published and have since become a part of the stock toolset in Houdini, which in turn allowed us to build more on top of Houdini’s built-in SOPs for Moana 2’s water FX workflow.

On the rendering side, the main challenge on the original Moana for rendering water was the need to combine water surfaces from many different sources (procedural, manually animated, and simulated) into a single seamless surface that could be rendered with proper refraction, internal volumetric effects, caustics, and so on. Our solution to combine different water surfaces on the original Moana was to convert all input water elements into signed distance fields, composite all of the signed distance fields together into a single world-spanning levelset, and then mesh that levelset into a triangle mesh for ray intersection [Palmer et al. 2017]. While this process produced great visual results, running this entire world-spanning levelset compositing and meshing operation at renderer startup for each frame proved to be completely untenable due to how slow it made interaction for artists, so an extensive system for pre-caching ocean surfaces overnight to disk had to be built out. All in all, the water rendering and caching system on the first Moana required a dedicated team of over half a dozen developers and TDs to maintain, and setting up the levelset compositing system correctly proved to be challenging for artists.

For Moana 2, we decided to revisit water rendering with the goal of coming up with something much easier for artists to use, much faster to render, and much easier to maintain by a smaller group of engineers and TDs. Over the course of about half a year, we completely replaced the old levelset compositing and meshing system with a new ray-intersection-time CSG system. Our new system requires almost zero work for artists to set up, requires zero preprocessing time before renderer startup and zero on-disk caching, renders with negligible impact on ray tracing speed, and required zero dedicated TDs and only part of my time as an engineer to support once primary development was completed. In addition to all of the above, the new system also allows for both better looking and more memory efficient water because the level of detail that water meshes have to exist at is no longer constrained by the resolution of a world-size meshed levelset, even with an adaptive levelset meshing. I think this was a great example where by revisiting a world that we already knew how to make, we were given an opportunity to reevaluate what we learned on Moana in order to come up with something better by every metric for Moana 2.

We knew that returning to the world of Moana was likely going to require a heavy lift from a volume rendering perspective. With a mind towards this, we worked closely with Disney Research Studios in Zürich to implement next-generation volume path guiding techniques in Hyperion, which wound up not seeing wide deployment this time but nonetheless proved to be a fun and interesting project from which we learned a lot. We also realized that the third act’s storm battle was going to be incredibly challenging from both an FX and rendering perspective. My last few months on Moana 2 were spent helping get the storm battle sequences finished; one extremely unusual thing we wound up doing was providing custom builds of Hyperion with specific optimizations tailored to the specific requirements of the storm sequence, sometimes going as far as to provide specific builds and settings tailored on a per-shot basis. Normally this is something any production rendering team tries to avoid if possible, but one of the benefits of having our own in-house team and our own in-house renderer is that we are still able to do this when the need arises. From a personal perspective, being able to point at specific shots and say “I wrote code for that specific thing” is pretty neat!

From both a story and a technical perspective, Moana 2 is everything we loved from Moana brought back, plus a lot of fun, big, bold new stuff. Making Moana 2 both gave us new challenges to solve and allowed us to revisit and come up with better solutions to old challenges from Moana. I’m incredibly proud of the work that my teammates and I were able to do on Moana 2; I’m sure we’ll have a lot more to share at SIGGRAPH 2025, and in the meantime I strongly encourage you to see Moana 2 on the biggest screen you can find!

To give you a taste of how beautiful this film looks, here are some frames from Moana 2, 100% created using Disney’s Hyperion Renderer by our amazing artists. These are presented in no particular order:

All images in this post are courtesy of and the property of Walt Disney Animation Studios.

References

Brent Burley, David Adler, Matt Jen-Yuan Chiang, Ralf Habel, Patrick Kelly, Peter Kutz, Yining Karl Li, and Daniel Teece. 2017. Recent Advancements in Disney’s Hyperion Renderer. In ACM SIGGRAPH 2017 Course Notes: Path Tracing in Production Part 1. 26-34.

Brent Burley, David Adler, Matt Jen-Yuan Chiang, Hank Driskill, Ralf Habel, Patrick Kelly, Peter Kutz, Yining Karl Li, and Daniel Teece. 2018. The Design and Evolution of Disney’s Hyperion Renderer. ACM Transactions on Graphics 37, 3 (Jul. 2018), Article 33.

Brent Burley. 2019. On Histogram-Preserving Blending for Randomized Texture Tiling. Journal of Computer Graphics Techniques 8, 4 (Nov. 2019), 31-53.

Brent Burley and Francisco Rodriguez. 2022. Fracture-Aware Tessellation of Subdivision Surfaces. In ACM SIGGRAPH 2022 Talks. Article 10.

Brent Burley, Brian Green, and Daniel Teece. 2024. Dynamic Screen Space Textures for Coherent Stylization. In ACM SIGGRAPH 2024 Talks. Article 50.

Matt Jen-Yuan Chiang, Peter Kutz, and Brent Burley. 2016. Practical and Controllable Subsurface Scattering for Production Path Tracing. In ACM SIGGRAPH 2016 Talks. Article 49.

Matt Jen-Yuan Chiang and Brent Burley. 2018. Plausible Iris Caustics and Limbal Arc Rendering. In ACM SIGGRAPH 2018 Talks, Article 15.

Matt Jen-Yuan Chiang, Yining Karl Li, and Brent Burley. 2019. Taming the Shadow Terminator. In ACM SIGGRAPH 2019 Talks. Article 71.

Henrik Dahlberg, David Adler, and Jeremy Newlin. 2019. Machine-Learning Denoising in Feature Film Production. In ACM SIGGRAPH 2019 Talks. Article 21.

Ralf Habel. 2017. Volume Rendering in Hyperion. In ACM SIGGRAPH 2017 Course Notes: Production Volume Rendering. 91-96.

Wei-Feng Wayne Huang, Peter Kutz, Yining Karl Li, and Matt Jen-Yuan Chiang. 2021. Unbiased Emission and Scattering Importance Sampling for Heterogeneous Volumes. In ACM SIGGRAPH 2021 Talks. Article 3.

Chenfafu Jiang, Craig Schroeder, Andrew Selle, Joseph Teran, and Alexey Stomakhin. 2015. The Affine Particle-in-Cell Method. ACM Transactions on Graphics (Proc. of SIGGRAPH) 34, 4 (Aug. 2015), Article 51.

Peter Kutz, Ralf Habel, Yining Karl Li, and Jan Novák. 2017. Spectral and Decomposition Tracking for Rendering Heterogeneous Volumes. ACM Transactions on Graphics (Proc. of SIGGRAPH) 36, 4 (Aug. 2017), Article 111.

Harmony M. Li, George Rieckenberg, Neelima Karanam, Emily Vo, and Kelsey Hurley. 2024. Optimizing Assets for Authoring and Consumption in USD. In ACM SIGGRAPH 2024 Talks. Article 30.

Yining Karl Li, Charlotte Zhu, Gregory Nichols, Peter Kutz, Wei-Feng Wayne Huang, David Adler, Brent Burley, and Daniel Teece. 2024. Cache Points for Production-Scale Occlusion-Aware Many-Lights Sampling and Volumetric Scattering. In Proc. of Digital Production Symposium (DigiPro 2024). 6:1-6:19.

Tad Miller, Harmony M. Li, Neelima Karanam, Nadim Sinno, and Todd Scopio. 2022. Making Encanto with USD: Rebuilding a Production Pipeline Working from Home. In ACM SIGGRAPH 2022 Talks. Article 12.

Thomas Müller. 2019. Practical Path Guiding in Production. In ACM SIGGRAPH 2019 Course Notes: Path Guiding in Production. 37-50.

Thomas Müller, Markus Gross, and Jan Novák. 2017. Practical Path Guiding for Efficient Light-Transport Simulation. Computer Graphics Forum (Proc. of Eurographics Symposium on Rendering) 36, 4 (Jun. 2017), 91-100.

Sean Palmer, Jonathan Garcia, Sara Drakeley, Patrick Kelly, and Ralf Habel. 2017. The Ocean and Water Pipeline of Disney’s Moana. In ACM SIGGRAPH 2017 Talks. Article 29.

Alexey Stomakhin and Andy Selle. 2017. Fluxed Animated Boundary Method. ACM Transactions on Graphics (Proc. of SIGGRAPH) 36, 4 (Aug. 2017), Article 68.

Emily Vo, George Rieckenberg, and Ernest Petti. 2023. Honing USD: Lessons Learned and Workflow Enhancements at Walt Disney Animation Studios. In ACM SIGGRAPH 2023 Talks. Article 13.

Thijs Vogels, Fabrice Rousselle, Brian McWilliams, Gerhard Röthlin, Alex Harvill, David Adler, Mark Meyer, and Jan Novák. 2018. Denoising with Kernel Prediction and Asymmetric Loss Functions. ACM Transactions on Graphics (Proc. of SIGGRAPH) 37, 4 (Aug. 2018), Article 124.

Tizian Zeltner, Brent Burley, and Matt Jen-Yuan Chiang. 2022. Practical Multiple-Scattering Sheen Using Linearly Transformed Cosines. In ACM SIGGRAPH 2022 Talks. Article 7.

Footnotes

¹ Our deep learning denoiser technology is one of the 2025 Academy of Motion Picture Arts and Sciences Scientific and Engineering Award winners. keyboard_return

This year at DigiPro 2024, we had a conference paper that presents a deep dive into Hyperion’s unique solution to the many-light sampling problem; we call this system “cache points”. DigiPro is one of my favorite computer graphics conferences precisely because of the emphasis the conference places on sharing how ideas work in the real world of production, and with this paper we’ve tried to combine a more traditional academic theory paper with DigiPro’s production-forward mindset. Instead of presenting some new thing that we’ve recently come up with and have maybe only used on one or two productions so far, this paper presents something that we’ve now actually had in the renderer and evolved for over a decade, and along with the core technique, the paper also goes into lessons we’ve learned from over a decade of production experience.

Here is the paper abstract:

A hallmark capability that defines a renderer as a production renderer is the ability to scale to handle scenes with extreme complexity, including complex illumination cast by a vast number of light sources. In this paper, we present Cache Points, the system used by Disney’s Hyperion Renderer to perform efficient unbiased importance sampling of direct illumination in scenes containing up to millions of light sources. Our cache points system includes a number of novel features. We build a spatial data structure over points that light sampling will occur from instead of over the lights themselves. We do online learning of occlusion and factor this into our importance sampling distribution. We also accelerate sampling in difficult volume scattering cases.

Over the past decade, our cache points system has seen extensive production usage on every feature film and animated short produced by Walt Disney Animation Studios, enabling artists to design lighting environments without concern for complexity. In this paper, we will survey how the cache points system is built, works, impacts production lighting and artist workflows, and factors into the future of production rendering at Disney Animation.

The paper and related materials can be found at:

• Project Page (Author’s Version and Supplementary Material)
• Official Print Version (ACM Library)

One extremely important thing that I tried to get across in the acknowledgements section of the paper and presentation and that I want to really emphasize here is: although I’m the lead author of this paper, I am not at all the lead developer or primary inventor of the cache points system. Over the past decade, many developers have since contributed to the system and the system has evolved significantly, but the core of cache points system was originally invented by Gregory Nichols and Peter Kutz, and the volume scattering extensions were primarily developed by Wei-Feng Wayne Huang. Since Greg, Peter, and Wayne are no longer at Disney Animation, Charlotte and I wound up spearheading the paper because we’re the developers who currently have the most experience working in the cache points system and therefore were in the best position to write about it.

The way this paper came about was somewhat circuitous and unplanned. This paper actually originated as a section in what was intended to have been a course at SIGGRAPH a few years ago on path guiding techniques, to have been presented by Intel’s graphics research group, Disney Research Studios, Disney Animation’s Hyperion team, WetaFX’s Manuka team, and Chaos Czech’s Corona team. However, because of scheduling and travel difficulties for several of the course presenters, the course wound up having to be withdrawn, and the material we had put together for presenting cache points got shelved. Then, as the DigiPro deadline started to approach this year, we were asked by higher ups in the studio if we had anything that could make a good DigiPro submission. After some thought, we realized that DigiPro was actually a great venue for presenting the cache points system because we could structure the paper and presentation as a combination of technical breakdown and production perspective from a decade’s worth of production usage. The final paper is a composed from three sources: a reworked version of what we had originally prepared for the abandoned course, a greatly expanded version of the material from our 2021 SIGGRAPH talk on our cache point based volume importance sampling techniques [Huang et al. 2021], and a bunch of new material consisting of production case studies and results on production scenes.

Overall I hope that the final paper is an interesting and useful read for anyone interested in light transport and production rendering, but I have to admit, I think that there are a couple of things I would have liked to rework and improve in the paper if we had more time. I think the largest missing piece from the paper is a direct head-to-head comparison with a light BVH approach [Estevez and Kulla 2018]; in the paper and presentation we discuss how our approach differs from light BVH approaches and why we chose our approach over a light BVH, but we don’t actually present any direct comparisons in the results. In the past we actually have more directly compared cache points to a light BVH implementation, but in the window we had to write this paper, we simply didn’t have enough time to resurrect that old test, bring it up to date with the latest code in the production renderer, and conduct a thorough performance comparison. Similarly, in the paper we mention that we actually implemented Vevoda et al. [2018]’s Bayesian online regression approach in Hyperion as a comparison point, but again, in the writing window for this paper, we just didn’t have time to put together a fair up-to-date performance comparison. I think that even without these comparisons our paper brings a lot of valuable information and insights (and evidently the paper referees agreed!), but I do think that the paper would be stronger had we found the time to include those direct comparisons. Hopefully at some point in the near future I can find time to do those direct comparisons as a followup and put out the results in a supplemental followup or something.

Another detail of the paper that sits in the back of my head for revisting is the fact that even though cache points provides correct unbiased results, a lot of the internal implementation details depend on essentially empirically derived properties. Nothing in cache points is totally arbitrary per se; in the paper we try to provide a strong rationale or justification for how we arrived upon each empirical property through logic and production experience. However, at least from an abstract mathematical perspective, the empirically derived stuff is nonetheless somewhat unsatisfying! On the other hand, however, in a great many ways this property is simply part of practical reality- what puts the production in production rendering.

A topic that I think would be a really interesting bit of future work is combining cache points with ReSTIR [Bitterli et al. 2020]. One of the interesting things we’ve found with ReSTIR is that in terms of absolute quality, ReSTIR generally can benefit significantly from higher quality initial input samples (as opposed to just uniform sampling), but the quality benefit is usually more than offset by the greatly increased cost of drawing better initial samples from something like a light BVH. Walking a light BVH on the GPU is a lot more computationally expensive than just drawing a uniform random number! One thought that I’ve had is that because cache points aren’t hierarchical, we could store them in a hash grid instead of a tree, allowing for fast constant-time lookups that might provide a better quality-vs-cost tradeoff that in turn might make use with ReSTIR feasible.

The presentation for this paper was an interesting challenge and a lot of fun to put together. Our paper is very much written with a core rendering audience in mind, but the presentation at the DigiPro conference had to be built for a more general audience because the audience at DigiPro includes a wide, diverse selection of people from all across computer graphics, animation, and VFX, with varying levels of technical background and varying levels of familiarity with rendering. The approach we took for the presentation was to keep things at a much higher level than the paper and try to convey the broad strokes of how cache points work and focus more on production results and lessons, while referring to the paper for the more nitty gritty details. We put a lot of work into including a lot of animations in the presentation to better illustrate how each step of cache points works; the way we used animations was directly inspired by Alexander Rath’s amazing SIGGRAPH 2023 presentation on Focal Path Guiding [Rath et al. 2023]. However, instead of building custom presentation software with a built-in 2D ray tracer like Alex did, I just made all of our animations the hard and dumb way in Keynote.

Another nice thing the presentation includes is a better visual presentation (and somewhat expanded version) of the paper’s results section. A recording of the presentation is available on both my project page for the paper and on the official Disney Animation website’s page for the paper. I am very grateful to Dayna Meltzer, Munira Tayabji, and Nick Cannon at Disney Animation for granting permission and making it possible for us to share the presentation recording publicly. The presentation is a bit on the long side (30 minutes), but hopefully is a useful and interesting watch!

References

Benedikt Bitterli, Chris Wyman, Matt Pharr, Peter Shirley, Aaron Lefohn, and Wojciech Jarosz. 2020. Spatiotemporal Reservoir Sampling for Real-Time Ray Tracing with Dynamic Direct Lighting. ACM Transactions on Graphics (Proc. of SIGGRAPH) 39, 4 (Jul. 2020), Article 148.

Alejandro Conty Estevez and Christopher Kulla. 2018. Importance Sampling of Many Lights with Adaptive Tree Splitting. Proc. of the ACM on Computer Graphics and Interactive Techniques (Proc. of High Performance Graphics) 1, 2 (Aug. 2018), Article 25.

Wei-Feng Wayne Huang, Peter Kutz, Yining Karl Li, and Matt Jen-Yuan Chiang. 2021. Unbiased Emission and Scattering Importance Sampling for Heterogeneous Volumes. In ACM SIGGRAPH 2021 Talks. Article 3.

Alexander Rath, Ömercan Yazici, and Philipp Slusallek. 2023. Focus Path Guiding for Light Transport Simulation. In ACM SIGGRAPH 2023 Conference Proceedings. Article 30.

Petr Vévoda, Ivo Kondapaneni, and Jaroslav Křivánek. 2018. Bayesian Online Regression for Adaptive Direct Illumination Sampling. ACM Transactions on Graphics (Proc. of SIGGRAPH) 37, 4 (Aug. 2018), Article 125.