DreamWorks Animation

DreamWorks Animation used AI for crowd simulation and character motion retargeting in Kung Fu Panda 4 (2024), allowing the studio to reuse complex movement data from earlier films and apply it to thousands of background characters with varied visual appearances in densely populated city scenes.

Secondary reporting indicates DreamWorks employed AI-driven crowd simulation on Kung Fu Panda 4 to populate large scenes — including Juniper City, described as densely packed with characters at different size ratios — by reusing and retargeting animation data from previous productions. This approach is described as part of a broader industry shift toward using AI to handle background character animation at scale. The studio's production team described the film as containing rich layers of movement among crowd characters. No primary source from DreamWorks specifically describes the AI tooling used for this purpose.

DreamWorks Animation deployed an internal machine learning model server called ForestFlow to manage the deployment and lifecycle of ML models used in its animation production pipeline, particularly for scheduling and optimizing rendering jobs. DreamWorks open-sourced ForestFlow in 2020 through the Linux Foundation AI project.

ForestFlow is a scalable, policy-based, cloud-native machine learning model server built by DreamWorks' principal architect Ahmad Alkilani to solve the challenge of moving ML models into production at scale. According to Alkilani, the system was developed because ML models affecting rendering job scheduling had a 'material impact' on the studio's bottom line, and keeping those models current as production data changed was a key operational need. ForestFlow supports automated model retraining triggers, A/B testing, canary deployments, and performance-based model routing. DreamWorks contributed ForestFlow to the LF AI Foundation as an open-source incubation project in April 2020.

DreamWorks Animation deployed MoonRay, its in-house production renderer, across all of its recent feature films including How to Train Your Dragon: The Hidden World, Kung Fu Panda 4, The Wild Robot, and The Bad Guys 2. MoonRay includes AI-powered denoising that cleans up visual noise in rendered images, allowing artists to see near-final quality results faster during production.

MoonRay is a Monte Carlo ray-tracing renderer developed internally at DreamWorks and used on all recent feature films. It incorporates GPU render denoising via Intel's Open Image Denoise, an AI-based image processing tool that removes noise artifacts from partially rendered frames, allowing artists to work with lower sample counts and faster iteration times. MoonRay was first used on the short film Bilby in 2018, made its feature debut on How to Train Your Dragon: The Hidden World in 2019, and was open-sourced by DreamWorks in March 2023 under an Apache 2.0 license. It remains in active internal use at DreamWorks and was used on The Wild Robot (2024) and The Bad Guys 2.