Holy Splats, Splatman! Excitement and support surrounding the free 3D engine Blender has continued to grow over the last month, with projects like Gaussian Frosting and Kiri Engine’s plugin. Now there is another extension into the world of Radiance Fields and it supports two methods of creating a Radiance Field, NeRFs and Gaussian Splatting. Both NeRFs and Gaussian Splatting are Radiance Fields.
There is a big shortage of datasets, both real and synthetic, but Splatman is here to greatly speed up the workflow of creating synthetic data for people to build ontop of. Splatman, created by Svenja Strobel and Colin Behrens, which are known as Vertex Wizards online, will help with generating the training data to be used in places like Nerfstudio, Postshot, or most recently Brush, but you will not be able to train them inside of Blender with the add on. If only there was a way to train Radiance Fields natively in Blender..
Splatman was designed to streamline these steps into a cohesive and user-friendly addon for popular open-source 3D software, Blender. The setup process is straightforward:
Scene Preparation: After installing Splatman, users begin by selecting an output directory and defining a bounding box around the core elements of their scene, isolating only the objects intended for reconstruction.
Smart Camera Placement: Depending on the scene type, users can place cameras for different environments. Options include Central Object for scenes with a singular focus, like a statue, or Large Scene for open settings like landscapes or interior rooms. Splatman auto-generates camera positions based on user-defined parameters for frame count and positioning.
Point Cloud Creation: With a single click, Splatman converts all scene meshes into a point cloud, ready for export. Users can adjust point density ranging from 100,000-1M points for optimal detail depending on their available memory and scene complexity. To help, it can be good to experiment with the density value. Typically, starting with - 4000 as an input is recommended.
Render and Export: Splatman configures the render settings automatically, making sure the resolution, rendering time, and output directories are optimized. After rendering the camera positions as an animation, users can export the necessary transforms JSON files, enabling seamless integration with NeRF and Gaussian Splatting frameworks.
To install Splatman, simply navigate to Edit > Preferences > Addons in Blender, select Install from File, and load the .zip file. Once installed, the settings can be accessed in Blender’s “n” panel on the right side.
Splatman offers a comprehensive suite of configurable settings that provide users with full control over the dataset creation process. Begin by selecting the export file path, which determines where all output files—such as images, point clouds, and camera alignments will be saved. It’s recommended to create a new folder for each project to keep everything organized. Users can then choose to either use the automatically calculated focal length or set a custom one, with a suggested minimum of 18mm.
Point cloud density can also be adjusted depending on the scene size, with a value of around 4000 being ideal for most cases. The rendering resolution is set to a default of 1000x1000 pixels, which ensures efficient performance without excessive resource use.
Splatman also allows for strategic camera placement based on the type of scene. For object-centric scenes, such as a character or a single object, the Create Camera Hemisphere / Central Object option places cameras around the focal point. For larger environments, like architectural visualizations or open nature scenes, the Create Camera Large Scene / Indoor option is more suitable. Users can adjust settings such as the minimum camera distance and minimum camera rotation distance to ensure optimal coverage. The addon even includes safety features like minimum tilt angle adjustments to avoid redundant captures when cameras are too close to objects.
Once the cameras are placed, users can generate a dense point cloud by clicking Create and Export Point Cloud, with the density adjustable based on project requirements. The generated point cloud will be automatically saved to the selected directory, ready for use. Splatman also offers a Prepare Render Settings feature that copies important settings—such as resolution and focal length—directly to the Blender scene, ensuring consistency throughout the rendering process.
After rendering the camera frames as an animation, users can finalize the process by exporting the camera alignments as JSON files with the Export Camera Files option. This step creates the transforms_train.json and transforms_test.json files, which are essential for training Radiance Fields, like NeRF and Gaussian Splatting.
As a free tool, Splatman brings high-level dataset creation capabilities within reach for artists and researchers of all levels. While support is limited, the developers welcome feedback and bug reports from the community, highlighting Splatman’s open-source roots. Users can also support ongoing development through donations.
Splatman can be downloaded for free from the Blender Market or for whatever amount a user would like to pay. Please note that at this time, Splatman is not designed for commercial use and redistribution, modification, or any commercial use of this addon is strictly prohibited. However, as more users begin using the plugin, this will likely shift.
