This article argues that the core stages of 3D animation, modelling through compositing, will remain stable over the next five years, but the implementation of each stage will shift as standards, real-time compute, and neural methods become routine. It positions OpenUSD as the organising layer that replaces monolithic scene files with scalable composition and asset graphs, with AOUSD’s ongoing releases signalling that interchange and variant management are becoming production defaults rather than specialist concerns [1,3]. It then outlines where neural techniques are already landing in everyday workflows: rapid capture of environments and props through Gaussian splatting for layout and lighting, and text-to-video systems accelerating early previs and pitch exploration while leaving final character performance dependent on editability and control. Finally, it connects hardware advances and real-time tool maturity to a narrowing gap between preview and final pixels, arguing that teams who invest in structure, auditability, and editable outputs will gain speed without sacrificing craft.
[1] Alliance for OpenUSD, “Announcing OpenUSD v25.11: Key Features and Improvements,” Nov. 12, 2025. Accessed: Nov. 20, 2025. The Alliance for OpenUSD (AOUSD)
[2] Alliance for OpenUSD, “Alliance for OpenUSD Announces New Members, Interest Groups, and Working Group Progress,” Mar. 17, 2025. Accessed: Nov. 20, 2025. The Alliance for OpenUSD (AOUSD)
[3] Pixar Animation Studios, “Official open source release of Universal Scene Description (USD),” Jul. 26, 2016. Accessed: Nov. 20, 2025. openusd.org
[4] NVIDIA, “NVIDIA Blackwell Platform Arrives to Power a New Era of Computing,” Mar. 18, 2024. Accessed: Nov. 20, 2025. NVIDIA Newsroom
[5] The Verge, “Nvidia announces next-gen RTX 5090 and RTX 5080 GPUs,” Jan. 6, 2025. Accessed: Nov. 20, 2025. The Verge
[6] AVNetwork, “Disguise is upping the real-time visual experience with the GX 3+,” Sept. 2025. Accessed: Nov. 20, 2025. AVNetwork
[7] SIGGRAPH Conferences, “Gaussian Splatting-based Rendering for High-quality 3D Content Creation,” Mar. 11, 2025. Accessed: Nov. 20, 2025. blog.siggraph.org
[8] C. Zeng et al., “RenderFormer: Transformer-based Neural Rendering of Triangle Scenes,” Microsoft Research, 2025. Accessed: Nov. 20, 2025. Microsoft
[9] Reuters, “OpenAI releases text-to-video model Sora for ChatGPT Plus and Pro users,” Dec. 9, 2024. Accessed: Nov. 20, 2025. Reuters
[10] The Guardian, “Tyler Perry halts $800m studio expansion after being shocked by AI,” Feb. 23, 2024. Accessed: Nov. 20, 2025. The Guardian
[11] AP News, “OpenAI releases AI video generator Sora but limits how it depicts people,” Dec. 2024. Accessed: Nov. 20, 2025. AP News
[12] E. Malakhatka et al., “XR Experience Design and Evaluation Framework,” book chapter, on interoperability and inclusivity layers. Accessed: Nov. 20, 2025.
[13] Business of Fashion and McKinsey, The State of Fashion 2025, on AI use cases and discovery. Accessed: Nov. 20, 2025.
For thirty years the basics of 3D animation have looked stable, a familiar sequence of modelling, rigging, keyframing, simulation, rendering and compositing. The tools and compute budgets have shifted, but the mental model has held. Over the next five years that model will still guide teams, however the implementation will change at almost every layer. Standards will mature, real-time and offline will blur, and neural methods will become normal rather than novel.
A typical 3D animation pipeline still proceeds through predictable phases:
• Modelling, the creation of geometry and materials.
• Rigging, the setup of controls and constraints that drive deformation.
• Animation, the performance itself, usually with a mix of keyframes, procedural tools and motion capture.
• Simulation, for secondary motion such as cloth, hair and effects.
• Lighting and rendering, where scenes become images.
• Compositing, the final assembly, cleanup and delivery.
These steps are the grammar of computer animation. Teams arrange them differently, and some phases may iterate in parallel, but the fundamentals persist across feature work, advertising, games and virtual production.
From the mid-1990s to the mid-2010s, advances clustered around three themes: better physically based rendering, broader GPU acceleration and incremental tool design inside large digital content creation packages. Pixar’s Universal Scene Description, released as open source in 2016, marked a new direction, shifting attention from monolithic scene files to scalable scene composition and interoperable asset graphs. openusd.org
In the 2020s, the pipeline began to look less like a single application and more like an ecosystem. The Alliance for OpenUSD, formed in 2023 by industry vendors and studios, continued to add members and working groups through 2024 and 2025, signalling that asset interoperability is no longer a special case, it is the default expectation. NVIDIA Newsroom+1 Recent AOUSD updates, including the v25.11 release in November 2025, reinforce that OpenUSD is on a regular cadence, with feature work landing across industries that rely on 3D data. The Alliance for OpenUSD (AOUSD)
On the hardware side, NVIDIA’s Blackwell architecture announcements in 2024, followed by consumer RTX 50-series launches in early 2025, brought large gains for real-time ray tracing and AI-assisted rendering, which benefit animation preview, look-development and in-camera visualisation. NVIDIA Newsroom+2NVIDIA+2 The practical impact is visible in adjacent tooling, for example new media servers for live and immersive shows that cite significant performance increases for complex generative content and real-time global illumination, the same kinds of workloads used in animated storytelling and previs. AVNetwork
Neural graphics techniques, once confined to research and tightly scoped production tests, are moving into the everyday tool set. Gaussian splatting has matured through 2024 and 2025 into practical pipelines for rapid capture of environments and props, feeding layout and lighting with high-fidelity proxies that can be refined or replaced as required. blog.siggraph.org Transformer-based neural renderers that operate on triangle scenes hint at faster path tracing alternatives that integrate more directly with traditional assets. Microsoft
At the same time, text-to-video systems changed expectations around previsualisation and motion studies. OpenAI’s Sora entered limited release in late 2024, with availability expanding through 2025, and its public demonstrations triggered prominent industry reactions about labour, authorship and safety. Reuters+2AP News+2 These models do not replace character animation, which still requires control, continuity and editability, however they have already shortened exploratory phases and pitch cycles. Teams can quickly trial camera language and tone, then translate selected beats into controllable 3D shots.
The policy and rights context is still unsettled. Reports and commentary across 2024 and 2025 document concerns about training data, watermarking and likeness protection, which studios must factor into deployment plans for generative tools. Brookings+2Wikipedia+2 The net effect is pragmatic adoption, with creative direction and production security acting as gates.
For studios and brands that need to move assets across film, interactive and spatial channels, the shift to OpenUSD-centred workflows reduces duplication. AOUSD governance updates in 2025, including new Interest Groups and working group progress, show that the standard is broadening to cover more day-to-day concerns, such as variant management, material models and live collaboration. The Alliance for OpenUSD (AOUSD) In parallel, XR design literature emphasises interoperability as a top-layer concern for production teams who intend to deploy content across devices and contexts.
For The VOLTAS readership, which spans digital fashion and spatial commerce, this matters. Digital product pipelines already depend on consistent asset schemas and predictable behaviour across engines. XR frameworks also underline inclusivity and accessibility as non-optional requirements, a reminder that the same content will be consumed by different bodies in different spaces.
Modelling will split between traditional polygonal workflows and capture-first approaches. Expect routine use of photogrammetry, neural radiance fields and Gaussian splats to establish starting geometry for sets and props, with procedural retopology and material baking bridging into engine-ready assets. Research directions in splatting and neural rendering point to stable improvements in quality and speed that align with production needs. blog.siggraph.org+1
Rigging will remain essential for characters, yet more deformation will be learned. We will see rig-assist tools that infer skinning weights, correctives and gait cycles from small motion libraries. Editors will expose learned deformation as simple sliders, not opaque black boxes, to preserve art-directability.
Animation will be more layered. Keyframes remain the spine of performance, but motion capture, simulation-aware solvers and generative motion will play larger supporting roles. Text-conditioned motion tools will assist with blocking and crowd behaviours, then animators will refine intent, rhythm and silhouette. Safety and provenance will be baked into review stages in response to policy and client requirements noted above. Brookings+1
Simulation will benefit from both hardware and method changes. Blackwell-class GPUs and successors raise practical limits for cloth, hair and destruction at previs and look-dev resolutions, reducing the historical gulf between real-time and offline previews. NVIDIA Newsroom+1
Lighting and rendering will converge across engines. Path tracers and real-time raster plus ray tracing hybrids now share material models and scene formats more reliably via USD. Expect more shows to adopt a single source of truth scene graph with branch targets for offline shots and in-camera or in-engine deliverables, rather than maintaining parallel scene authoring paths. AOUSD’s continued cadence through 2025 supports this direction. The Alliance for OpenUSD (AOUSD)
Compositing will incorporate more metadata and automation. USD-native exports with per-primitive IDs, AOVs and render-context tags will help comp teams automate mattes and relighting. In virtual production, editorial and comp will rely more on camera-tracked plate synthesis from neural captures, lowering reshoot pressure for missed inserts.
Core skills still matter. Artists who understand spacing, timing, arcs and weight will adapt to any tool. That said, hiring profiles are broadening. Studios increasingly ask for USD literacy, engine experience, and the ability to operate across DCCs without friction. Some pipeline roles now span schema design, data governance and review automation more than script-only tasks.
Education will follow. Blender’s geometry and simulation nodes, Autodesk’s USD tooling and engine-integrated layout will be used earlier in curricula. Peer sectors such as fashion that already navigate heavy product data and omnichannel delivery provide useful case material for animation teams working toward asset reuse and digital twins. The latest State of Fashion analysis points to generative tooling for discovery and personalisation, which mirrors how animation groups are trialling AI assistants inside production, with a focus on measurable impact rather than hype.
The basics of 3D animation are resilient. The sequence from modelling to compositing will still anchor production in 2030. What will change is everything around those steps, from asset graphs and interchange to how motion is proposed, reviewed and approved. Standards such as OpenUSD, stronger real-time hardware and pragmatic neural tools will not replace the craft, they will shorten feedback loops and move decisions earlier in the schedule. Teams that invest in structure, auditability and editability will be well placed to benefit from these shifts without losing the qualities that make animated work memorable.