Exploring the Potential of Neural Rendering in Visual EffectsNeural Radiance Fields (NeRFs) and Gaussian Splatting are starting to change the way visual effects artists capture, render, and integrate photorealistic scenes, especially environments. In this fxpodcast, Sam Hodge, a machine learning software engineer currently at the Australian Institute for Machine Learning (AIML), University of Adelaide, discusses these emerging techniques and their growing role in virtual production, previsualization, and postvisualization.From Photogrammetry to Neural Radiance FieldsNeRFs offer a major evolution from traditional photogrammetry and LiDAR scanning. While photogrammetry reconstructs static 3D models with baked-in lighting, NeRFs allow for view-dependent rendering, meaning elements like specular highlights and transparency shift naturally based on the cameras perspective. This enables virtual cameras to move through a scene with realistic light transport, a key benefit for VFX workflows.The scene Sam discusses and training in NeRF StudioNeRFs in Production: Strengths and ChallengesNeRFs capture complex materials and lighting conditions with an unprecedented level of realism. However, rendering them remains computationally intensive, with each frame requiring billions of calculations per pixel. This challenge makes real-time applications difficult. To address this limitation, Gaussian Splatting (GS) has emerged as a faster alternative, trading some accuracy for major speed gains.Both NeRFs and GS are not quite ready for high end tentpole final pixel VFX pipelines, but with enormous amounts of research being published on both, it is not unreasonable to expect this to change soon. Certainly they do have a role in the wider range of AI and ML tools that are appearing and fore projects aiming for non-theatrical exhibition.Gaussian Splatting: A Faster AlternativeGaussian Splatting reconstructs 3D environments faster than NeRFs while maintaining photorealistic quality. Instead of relying on polygonal meshes, it uses elliptical points that blend together to create smooth, volumetric visuals. While it does not achieve the same level of fine detail as NeRFs, they can run in real-time, making it ideal for virtual production, game engines, and immersive AR/VR experiences.Dynamic Scenes and 4D Gaussian SplattingNeRFs traditionally struggle with moving objects, but 4D Gaussian Splatting processes time-based data, allowing for the volumetric reconstruction of dynamic performances. This makes it possible to capture live-action events, such as a tennis match or a stage performance, and view them from infinite angles. As a result, 4D Gaussian Splatting opens new possibilities for immersive media, volumetric filmmaking, and interactive experiences.Integrating Neural Rendering into VFX PipelinesVFX professionals continue exploring ways to integrate these technologies into existing pipelines. NeRFs and Gaussian Splatting do not inherently produce polygonal meshes. In the case of NeRFs they live inside a neural network, for GS, they can be thought of as a point cloud.While not immediately providing polygon solutions, they so prove valuable opportunities for: Previsualization and postvisualization, creating photorealistic environments and testing camera movements. Virtual production, providing interactive, high-fidelity backdrops on LED volumes. Location scouting, enabling filmmakers to explore digital representations of real-world locations. Performance capture and digital human references, offering new ways to train AI-driven facial and motion models.A New Era for Machine Learning in VFXThis weeks fxpodcast discussion also highlights how machine learning is reshaping the industry, particularly in generating high-quality synthetic training data. By using NeRFs and Gaussian Splatting to capture real-world actors, environments, and materials, artists can train AI models without relying on scraped internet content. This approach ensures ethical and controlled data acquisition while improving AI-based workflows.A Tool, Not a FadNeRFs and Gaussian Splatting are not passing trends. As these technologies continue to evolve, they offer real advantages for environment creation, camera planning, and immersive media. While the industry is still refining their applications, they already prove their value in production.For a deeper look at these cutting-edge technologies, listen to the full fxpodcast with Sam Hodge.https://www.fxguide.com/wp-content/uploads/2025/03/Apple_NeRF.mp4Here are the data files that Sam and Mike discuss in the podcast:https://r2.fxguide.com/articles/2025-02/prague_video_filtered.ziphttps://r2.fxguide.com/articles/2025-02/PragueInstallerImage.dmg.zipnote: you can also load the project into NeRF Studio as we have above.