NeRF datasets from Blender, minus the scripting headache

What it does

BlenderNeRF is a Blender add-on that automates synthetic dataset creation for Neural Radiance Fields and Gaussian Splatting. It renders frames and exports camera parameters into transforms_train.json and transforms_test.json, packaging everything into a ZIP. The author built it to replace the “cumbersome code” usually needed to extract camera data from Blender for neural rendering.

The interesting bit

Instead of treating neural rendering as an afterthought, the add-on bakes dataset generation into Blender’s UI with three distinct sampling strategies: frame subsets for animations, separate train and test cameras, and randomized views sampled from a sphere. That flexibility lets you tailor synthetic data to how the model will actually be evaluated.

Key highlights

• Three export modes: Subset of Frames for camera animations, Train and Test Cameras for split setups, and Camera on Sphere for randomized spherical views
• Exports both standard NeRF and Instant NGP JSON formats, including the AABB scale parameter
• Optional Gaussian Splatting support via a points3d.ply point-cloud export generated from visible mesh vertices
• Packages renders, camera poses, and optional logs into a single ZIP file
• Runs entirely inside Blender 4.0+ as a panel in the 3D viewport

Caveats

• The author describes it as a “fun side project” developed mainly on macOS, so your mileage may vary on other platforms
• Creating objects manually named BlenderNeRF Sphere or BlenderNeRF Camera will break the add-on
• Gaussian Splatting workflows may need a dummy test camera file if you do not render separate test images

Verdict

A handy tool for VFX artists, researchers, or hobbyists who already work in Blender and want synthetic NeRF or Gaussian Splatting data without writing custom export scripts. Skip it if you are only interested in real-world scene reconstruction.