EN

Training Manual for Trainers: Scanning Spaces Using Gaussian Splatting | CULTUURCONNECT

Objective
Guide trainers in teaching how to scan spaces using Gaussian splatting to create highly detailed 3D models from pixel data using Luma.AI, emphasizing real-world application in digital arts and performance contexts. 

 

Structure of the Training Session 

  1. Preparation (approx. 10 minutes)
  • Objective: Set up the environment and tools for an efficient session. 
  • Steps: 
  1. Select the space to be scanned (indoor/outdoor with varied surfaces). 
  1. Prepare scanning equipment. 
  1. Install necessary software (e.g. Nerfstudio, Luma.AI, COLMAP). In this manual we will focus on Luma.AI. 
  1. Test file input/output compatibility for your chosen platform.
     

Checklist: 

  • All software installed and functioning 
  • Camera/device tested with sample capture 
  • Backup storage ready for large image datasets 
  • Basic walkthrough presentation prepared
     

 

  1. Introduction to Gaussian Splatting (approx. 15 minutes)
  • Objective: Introduce the technique and its creative potential. 
  • Steps: 
  1. Define Gaussian splatting in simple terms: “projecting colored 3D Gaussian blobs into space using 2D images to reconstruct detailed 3D geometry.” 
  1. Show examples comparing Gaussian splatting with traditional mesh/point-cloud scans. 
  1. Explain the options and possibilities of how to make a Gaussian splat (local Gaussian splatting vs. other less technical methods e.g. Luma.AI) 
  1. Emphasize its advantages: speed, finer detail, more natural lighting and responsive to virtual light sources.
     

Trainer Tip: Use visual comparisons – real object, mesh scan, Gaussian splat result. 

 

  1. Hands-On Capture and Processing (approx. 30 minutes)
  • Objective: Use Luma.AI to quickly create and visualize a 3D Gaussian splat without technical setup. 
  • Steps: 
  1. Download and open the Luma.AI app on an iPhone. 
  1. Capture the scene: Use the app to record photos or a video while walking around the object or environment. 
  1. Automatic upload and processing: Luma.AI uploads the images and begins training the model automatically. 
  1. View the model: You’ll get a notification once the model is ready. View it directly in the app or via a web browser. 
  1. Export the Gaussian splat as a .ply file for use in 3D engines or editing software.
     

Trainer Tip: Use this method for fast results and lower the technical barrier for beginners. Pre-load example scenes to show during the session. 

 

  1. Advanced Use and Troubleshooting (approx. 15 minutes)
  • Objective: Explore optimization and creative application. 
  • Steps: 
  1. Discuss tips for better detail (camera path planning, lighting). 
  1. Introduce downsampling vs. full-resolution trade-offs. 
  1. Show how to integrate Gaussian splat models into game engines or XR tools. 

 

  1. Wrap-Up and Feedback (approx. 10 minutes)
  • Objective: Reinforce key learnings and collect feedback. 
  • Steps: 
  1. Recap: Capture > Process > Render > Apply 
  1. Share links to open-source tools, tutorials, and sample data 
  1. Gather participant feedback on clarity and practical use 

 

Post-Training Follow-Up 

  • Provide cheat sheet of steps and commands 
  • Offer optional office hours for Q&A 
  • Encourage participants to scan their own space and share results 

Trainer Tip: Create a shared gallery or folder for participants to upload their models.