The Nerfstudio team has been steadily pushing the envelope on radiance field tooling and their latest update is a big one for anyone experimenting with Gaussian Splatting. A new pull request brings a major overhaul to the viewer and rendering pipeline, updating the older nerfview system to be more powerful, introducing a dedicated GsplatViewer, purpose-built for visualizing splats with more control and flexibility.

This update features the introduction of GsplatViewer, along with a supporting UI state class called GsplatRenderTabState. Together, they offer a reimagined viewer tailored specifically for splats, giving users much more freedom over how scenes are rendered. Instead of being locked into basic visualization modes, you can now easily toggle between RGB, depth, and alpha render modes, apply colormaps for perceptual depth clarity, and select from different camera models to better match real-world lenses. Visual quality also gets a boost with support for anti-aliasing, helping the rendered splats look cleaner without requiring extra training overhead.

The default example scripts like simple_trainer.py and simple_viewer.py now use GsplatViewer out of the box, aligning them with the updated rendering pipeline. The team also took the opportunity to remove leftover dependencies tied to the older nerfview implementation and to consolidate rendering parameters. Now, rendering supports consistent arguments like rasterize_mode and camera_model, making it easier to reason about how your data is being displayed.

Ultimately this means a smoother, more intuitive experience for anyone prototyping with Gaussian Splatting in Nerfstudio. The upgraded viewer makes it easier to experiment, iterate, and debug without worrying about mismatched viewer settings or legacy code paths.

Nerfstudio uses Viser as its web viewer, but for those who would like to only use gsplat, they can now benefit from the full Nerfstudio viewer interface.

Nerfstudio remains one of the most critical platforms in the radiance field ecosystem and powers several of the existing implementations. It is important that Nerfstudio continues to develop and be supported by the community at large. It remains open sourced and Apache 2.0 licensed for those who are considering contributing back to the project.