Foundry has released Nuke 17.1 in open beta, extending the native 3D Gaussian splatting toolset it introduced in 17.0 with support for dynamic or 4D splats, .splat file export, basic relighting through SplatRender, and an upgrade of the compositor’s 3D viewer to Hydra 2.0. The beta is publicly available across Nuke, NukeX, and Nuke Studio.
The animated splat workflow is the substantive addition, and it closes a specific gap. Nuke 17.0 made it possible to import, manipulate, and render static 3DGS data natively, reading .ply and .splat files through the revamped GeoImport and GeoReference nodes, color-correcting and editing point sets with GeoGrade and GeoDeletePoints, and rendering radiance fields back to 2D pixels with the new SplatRender node, as our earlier coverage of the 17.0 beta and the shipped 17.0 release detailed. What 17.0 did not handle was time. Splats were single frame reconstructions of real world locations used as environments or set extensions, with no path for sequences that change over the shot.
17.1 introduces a new GeoSequencer node to address this. GeoSequencer processes a sequence of .USD, .USDC, .PLY, or .splat files and outputs them as a time sampled .USDC file that Nuke’s existing splat toolset can then manipulate. Arbitrary animated attributes are detected in the input sequence and can be optionally passed through to the generated USD Value Clip, so per frame data such as normals or position can be carried downstream and rendered via SplatRender. The Value Clip route should be more performant than reading .ply files directly, since USDC files read faster and can be more memory efficient, with attribute data loaded as required rather than all at once.
It is also now possible to write .splat files out of the GeoExport node, alongside the PLY, USD, and ABC formats Nuke already supported, so an artist can bring a splat sequence into Nuke, work on it, and output it in the format the rest of the pipeline expects. The GeoExport node also defaults to a new flattened export workflow in which frames are written individually and then stitched into a single file.
SplatRender has been updated to support 2D rendering of Direct, Point, and Spot lights placed in Nuke’s 3D system, which the notes describe as “basic relighting of Gaussian splat scenes and elements for better integration into your comps.” This is a Nuke feature in its own right, distinct from the gaussian splat shading recently added to Autodesk’s Arnold. Foundry is explicit that normals and position data are not required for relighting, but that where they exist in the splat they can be rendered through SplatRender for use elsewhere in the comp. Control lives in an Enable Relighting toggle, a Lights selector defaulting to all Direct, Point, and Spot lights, an Ambient Color for areas no light reaches, and a Lighting Blur Radius that smooths the lit result in screen space while keeping object edges sharp. The DirectLight, PointLight, and SpotLight nodes gain a Shadow control set covering Strength, Softness, Depth Range, Onset, Bias, Resolutions, Depth Quality, and Cutoff.
The 3D viewer moves to Hydra 2.0, the current version of the USD rendering framework. Beyond refined viewer controls, the upgrade lets the viewer node set a Display AOV so a 3D scene can be previewed as individual channels such as depth and position under the HdStorm render delegate. This should 2.0 begin to clear the way for third party Hydra delegates in Nuke in future. The release also adds non-destructive USD export that can explicitly define sublayers and write scene modifications as sparse “over” overrides.
Foundry continues to pursue gaussian splatting in the core product, now with making them an animatable, lightable, exportable element that lives inside Nuke’s native 3D graph pipeline. It is available now in open beta.
