Gaussian Splat from Insta360: Step-by-Step Guide

Updated 14 November 2026 · Standard formula, shown below

This guide walks you through creating a 3D Gaussian splat—a photorealistic, viewable-from-any-angle scene—using footage from an Insta360 camera. You’ll learn how to capture suitable video, extract frames, and process them into a splat using free, widely available tools. The result is a shareable 3D model that can be viewed in a browser or VR headset.

What You’ll Need

An Insta360 camera (any model that shoots 360° video, e.g., X3, X4, ONE RS 1-Inch 360)
A computer with a dedicated GPU (NVIDIA recommended) and at least 16GB RAM
Insta360 Studio (free desktop app) or the mobile app for exporting frames
Postshot (free trial available) or Nerfstudio (open-source) for splat generation
A tripod or stable mount for the camera
Good lighting and a static scene (no moving objects)

Step-by-Step

Plan your capture: Choose a scene that is well-lit and has texture. Avoid reflective surfaces, transparent objects, and moving elements like people or cars. The area should be roughly 3–10 meters across for best results.
Set up the camera: Place the Insta360 on a tripod at chest height. Ensure the camera is level and the lens is clean. Set the resolution to the highest available (e.g., 5.7K or 8K) and frame rate to 30fps. Disable any stabilization features like FlowState, as they can alter the optical path.
Record the video: Start recording and slowly walk around the scene in a full circle, keeping the camera pointed toward the center. Maintain a steady pace and keep the camera at a constant distance from the subject. Overlap your path by about 20% to ensure complete coverage. Record for 1–2 minutes.
Export frames: Transfer the video to your computer. Open Insta360 Studio, import the clip, and use the frame export feature to save every 2nd or 3rd frame as a JPEG or PNG. Aim for 100–300 frames. Name them sequentially (e.g., frame_0001.jpg).
Prepare the frames: If using Nerfstudio, create a folder named images and place all frames inside. If using Postshot, you can import the folder directly. Ensure all frames are the same resolution and orientation.
Install the splatting tool: For Postshot, download and install the software from the official site. For Nerfstudio, follow the installation instructions for your OS (requires Python and CUDA).
Run the splatting process: In Postshot, create a new project, select the images folder, and click “Train.” In Nerfstudio, use the command ns-train gaussian-splatting --data /path/to/images. The process will take 10–60 minutes depending on your GPU.
Export the splat: Once training completes, export the model. Postshot exports a .ply file. Nerfstudio exports a .ply or .splat file. The .ply file can be viewed in most splat viewers.
View and share: Open the exported file in a viewer like SuperSplat or the Postshot viewer. You can also convert the .ply to a .splat for use in web viewers. Share the file or embed it in a webpage.

Common Mistakes

Moving objects: People, leaves, or cars in the scene will appear as blurry artifacts. Ensure the scene is completely static.
Poor lighting: Low light or harsh shadows reduce feature matching. Use diffuse, even lighting.
Too few frames: Less than 50 frames often results in a sparse splat with holes. Aim for at least 100.
Camera shake: Even with a tripod, walking can cause vibration. Use a remote trigger or record while standing still and rotating the camera on a gimbal.
Incorrect frame extraction: Exporting every frame can cause motion blur; exporting too few loses overlap. Use every 2nd or 3rd frame.

Tips

Use a 360 camera to capture the entire environment, but for a single object, you can crop the video to a 180° field of view before exporting frames.
For large scenes, capture multiple passes at different heights (e.g., ground level and chest height) to improve geometry.
Use a color calibration chart in the scene to ensure consistent white balance.
If using Nerfstudio, enable the --pipeline.datamanager.camera-optimizer flag to correct for slight camera misalignment.
Postshot offers a one-click optimization for handheld captures—use it if you didn’t use a tripod.

Troubleshooting

Training fails with CUDA out of memory: Reduce the number of frames or lower the resolution (e.g., 1024px). Close other GPU applications.
Splat looks blurry or has holes: Increase the number of frames, ensure more overlap, and check that the scene is well-textured. Try adding more viewpoints.
Frames are not aligned: If the camera moved too fast, motion blur occurs. Re-record with slower movement. In Postshot, enable “Optimize camera poses” during training.
Exported .ply is too large: Use the compression option in Postshot or decimate the point cloud in a tool like CloudCompare before exporting.
Viewer shows black screen: Update your graphics drivers and ensure the viewer supports .ply format. Convert to .splat if needed.

Frequently Asked Questions

Can I use any Insta360 camera?

Yes, any model that shoots 360° video works. Higher resolution (5.7K or above) yields better results.

Do I need a powerful PC?

A dedicated GPU with at least 8GB VRAM is recommended. Without it, training will be very slow or may fail.

How long does the process take?

Capture takes 2–5 minutes, frame extraction 5 minutes, and training 10–60 minutes depending on hardware.

Can I create a splat from a single photo?

No, you need multiple overlapping images from different angles. A 360 video provides those frames.

What is the best file format for sharing?

The .ply format is widely supported. For web viewing, convert to .splat using tools like the SuperSplat converter.

Can I edit the splat after creation?

Yes, you can remove unwanted points in editors like Postshot’s editing mode or CloudCompare.

Method: the standard formula for this calculation, rounded as each result states. See how the calculators work or report an error.