# Pipeline: Capture to Splat

## Introduction

This page walks the pipeline from raw footage to a trained splat model: capture planning, frame extraction, the camera solve, the training run, and cleanup. Each stage sets limits the next one cannot undo.

### Step 1: Capture Planning
- Overlap requirements (minimum 70-80%)
- Lighting and exposure considerations
- Avoiding motion blur and rolling shutter
- Video vs. stills — trade-offs

### Step 2: Frame Extraction
- Extracting frames from video
- Choosing frame interval (temporal vs. oversampling)
- Frame quality filtering

### Step 3: Camera Solving (SfM)
- COLMAP: sequential vs. exhaustive matching
- Importing poses from Metashape
- What makes a good SfM result for 3DGS

### Step 4: Training the Splat
- Postshot: user-friendly, GUI-based
- Brush: command-line, more control
- SkySplat: all-in-Blender pipeline
- Training parameters (iterations, resolution, pruning)

### Step 5: Cleanup and Export
- Removing floaters and background artifacts
- Cropping to region of interest
- Output formats (.ply, .splat)

## Summary

The pipeline is a chain of decisions, not a button press. Capture overlap, frame extraction, the camera solve, the training run, and the cleanup each set limits the next stage cannot undo. You now know the five steps and what each one costs the result.

<div className="closing-grid">

<div>

#### Resources

### KB Pages
- [Postshot tool page](../../../tools/postshot.md)
- [COLMAP tool page](../../../tools/colmap.md)
- [SkySplat addon page](../../../addons/skysplat.mdx)
- [KIRI 3DGS Render addon page](../../../addons/kiri-3dgs.mdx)
- [Course Syllabus](../syllabus.md) — full course overview and module map

</div>

<div>

#### Sources

<!-- 1. [Title — link](https://…) — one-line note <span id="ref-N" /> -->

</div>

</div>
