Nuke 17.1 Makes Gaussian Splats More Useful in Compositing
Gaussian splats have spent the past year looking like the next big thing in visual effects, which is usually the point when a production team starts asking whether the thing can survive a real shot. Nuke 17.1, released on August 20, 2026, gives the format a more useful answer. Foundry’s update adds animated splat sequences, basic relighting, broader USD support, and a cleaner route back into the wider pipeline.
That matters because a scan is easy to admire and harder to integrate. A static environment can sit behind a plate. An animated splat has to move with the shot, respond to the scene, and hold up when a client asks for a different mood at 4:47 p.m. Nuke 17.1 doesn’t turn Gaussian splats into conventional 3D assets, but it makes them more workable inside a compositor’s existing habits.
What changed in Nuke 17.1
The headline feature is the new GeoSequencer node. It processes sequences of USD, USDC, PLY, or SPLAT files and outputs a time-sampled USDC file for Nuke’s existing splat tools. Foundry says the generated value clip can be faster and more memory-efficient than reading PLY files directly because attribute data loads as required rather than all at once. The full details are in the Nuke 17.1v1 release notes.
The update also adds .splat export through GeoExport. That small detail is more important than it sounds. A compositing team can adjust a splat sequence in Nuke and hand it back to another part of the pipeline in a format that tool can actually read. Fewer one-way experiments, fewer mystery folders called “final_final_splat2.”
SplatRender now supports 2D rendering of Direct, Point, and Spot lights placed in Nuke’s 3D system. Relighting is controlled with options for enabling the effect, choosing scene lights, setting an ambient color, and adding a lighting blur radius. The same release adds shadow controls such as strength, softness, depth range, onset, bias, resolution, depth quality, and cutoff.
Relighting is useful, with an asterisk
Foundry describes the feature as basic relighting, and that wording is worth keeping. The goal is better integration into a comp, not a magical conversion of captured appearance into a physically based material. You can push a light across a splat environment, shape the shadow response, and make an element sit closer to the plate. You shouldn’t expect it to behave like a clean mesh with carefully authored normals, UVs, and physically separated textures.
That limitation is also a reasonable design constraint. Gaussian splats capture how a place looked from sampled viewpoints. They aren't a conventional asset waiting for a lighting pass. Nuke’s approach gives artists a controllable image result while preserving the speed of a compositor-led workflow. When the shot needs a slight key-light shift, a cooler fill, or a shadow pass that helps an element belong in the frame, that can be enough.
If a splat contains normals or position data, SplatRender can pass those channels out for other compositing work. That opens a more familiar compromise: use the built-in relighting for the broad adjustment, then use depth, position, or normals to refine the result with the tools already in the script.
Why animated splats change the workflow
Static splats are handy for set extensions, matte paintings, location reconstruction, and the occasional “we absolutely cannot go back to that location” conversation. Animated splats expand the category. A moving capture can become an environment element, a camera move can travel through a scanned space, and a generated sequence can live in a shot instead of serving as a one-frame proof of concept.
The practical workflow is fairly direct:
- Gather the sequence. Start with USD, USDC, PLY, or SPLAT files and confirm the frame range before opening the node graph. Missing frames are still missing frames, even when the format has a futuristic name.
- Build a value clip. Use GeoSequencer to generate a time-sampled USDC sequence. Select the animated attributes you need, such as position or normals, and let Nuke create a GeoImport node after generation.
- Place it in the shot. Use Nuke’s 3D system to transform the splat, line up the camera, and check the element at working resolution before adding expensive finishing work.
- Relight with restraint. Add Direct, Point, or Spot lights, then adjust ambient color and shadow settings until the element supports the plate. If the splat starts announcing its technical limitations, believe it.
- Render and hand off. Use SplatRender outputs for the comp, and export a .splat file from GeoExport when another department needs the manipulated sequence.
Hydra 2.0 makes the update bigger than splats
Nuke 17.1 also updates the 3D viewer to Hydra 2.0. Artists can assign a Hydra delegate in the viewer, preview display AOVs such as depth and position, and adjust delegate render settings without leaving the viewer. The new GeoRender node can output delegate passes directly into Nuke.
For studios moving between compositing, look development, and layout, the important change is less about a shiny viewport and more about keeping scene information intact. The release adds non-destructive USD export workflows, including explicit sublayers and sparse overrides. That reduces the pressure to flatten every adjustment into a file that the next person has to politely dismantle.
CG Channel’s overview of Nuke 17.1 notes support for delegates including RenderMan and MoonRay in the Hydra-based viewer, along with the new GeoRender workflow. The exact value will depend on a studio’s existing USD setup, but the direction is clear: Nuke is trying to act less like a final stop and more like a participant in a shared 3D scene.
What design and creative teams should do now
Don’t rewrite a studio pipeline around splats because a release note contains the word “animated.” Start with one contained shot. A location extension, product reveal, or title sequence is enough. Test camera alignment, render time, memory use, light response, and the handoff format. Keep a conventional fallback for anything that has to survive late client changes.
It’s also worth separating the creative brief from the technical experiment. A short brief from pixelbrief.ai can help define the visual target, lighting direction, palette, and failure conditions before the team spends an afternoon tuning a scene that was never meant to be in the final cut. The splat is a means to an image, not a personality.
For teams already using Nuke 17.0’s native Gaussian Splat tools, 17.1 is a meaningful production update. GeoSequencer makes time-based work more manageable, SplatRender gives artists a practical way to seat captured environments in a comp, and Hydra 2.0 strengthens the USD connection around them. The release doesn’t remove the format’s quirks. It does give those quirks a place to work.