📋 Workflows: from Capture to Export

A complete 3D capture workflow — from your first smartphone shot to a controlled pro setup.

Whether photogrammetry (photo → mesh) or Gaussian Splatting (photo/video → radiance field): the process is always the same — capture → processing → cleanup → export. This page walks you once through all four phases and shows at each step what beginners, intermediates and pros specifically do differently. All information is current as of 2026.

1. Which path suits you?

Before you start: there is no single workflow. Choose your entry point based on your gear and ambition. You can move up a level at any time later — the basic capture principles stay the same.

2. Phase 1 — Capture BeginnerIntermediatePro

Capture determines 90% of the final result. No software can rescue blurry photos, missing angles or changing light. The good news: the rules for photogrammetry and Gaussian Splatting are almost identical.

The shared foundation

• Overlap 70–80%: every photo must overlap heavily with the next one. Rule of thumb: one image every 10–15° — so 24–36 shots per rotation on a turntable, more for complex objects.
• Diffuse, constant light: overcast sky or softboxes. Otherwise hard shadows "bake" themselves into the texture/splat.
• Everything sharp: the entire object must be in focus in every image.
• The object must not change: no swaying leaves, no moving person, no changing light.

Camera settings Intermediate

Capture method

• Turntable (object rotates, camera fixed): ideal for small objects. But: the background must be neutral and must be masked out during alignment (see pitfalls).
• Walk-around (you move around the object): for large objects and scenes. Do two or three height rings (top/middle/bottom).
• Drone: for buildings/terrain. Grid flight with 80% front and 70% side overlap.

Specifically for Gaussian Splatting Intermediate

3DGS also handles video. Recommendation: 4K, 30 fps, very slow and steady movement so no motion blur occurs. Frames are extracted from the video later (Postshot & co. do this automatically). More important than in photogrammetry: absolutely constant lighting — splatting stores view-dependent reflections, and changing light produces "ghosts".

Pro techniques Pro

• Cross-polarization for glossy/smooth objects: a polarizing filter on both the light source and the lens, rotated against each other until the reflections disappear. Result: a matte, "raw" surface that the software reconstructs cleanly.
• Color chart + scale (e.g. ColorChecker, scale bar/coded targets) placed in the shot → correct colors and true scale.
• Scanning spray (a matting, sublimating spray) for reflective/transparent parts when polarization isn't enough.

3. Phase 2 — Processing BeginnerIntermediatePro

Now the software does the computing. The process differs depending on the target technique.

Photogrammetry pipeline

1. Import all images (develop RAW beforehand if needed, but without aggressive editing).
2. Alignment / Structure-from-Motion: the software finds common features and computes the camera position of each photo + a sparse point cloud. If this fails, something is wrong with overlap, sharpness or texture.
3. Dense cloud / depth maps: a dense point cloud.
4. Meshing: the points become a closed surface.
5. Texturing: the original photos are projected onto the mesh.

Gaussian Splatting pipeline

1. Frame extraction (for video) → individual images.
2. Camera poses via SfM: COLMAP computes where each image was taken from. GUI tools like Postshot handle this internally.
3. Training: millions of Gaussians are iteratively optimized (typically 15–30 min on a current NVIDIA GPU, 30k training steps as a common guideline).

4. Phase 3 — Cleanup IntermediatePro

A raw result ≠ a finished asset. This is where hobby separates from pro.

Mesh cleanup (photogrammetry)

• Remove background & turntable: cut away excess geometry.
• Close holes: in Blender (Mesh > Fill Holes) or directly in the photogrammetry software.
• Decimate: a raw scan quickly has millions of polygons. The Decimate modifier reduces it for further processing.
• Retopology Pro: for games/animation, lay a clean, even topology over the high-poly (e.g. with Instant Meshes / QuadRemesher).
• Texture baking Pro: bake color and normals from the high-poly onto the low-poly. This preserves detail at low polygon counts.

Splat cleanup (Gaussian Splatting)

• Remove floaters/"floating particles": the most important step. In SuperSplat (free, browser-based) select them with box/sphere/brush and delete.
• Crop the scene: limit it to the actual object, remove the surroundings.
• Adjust attributes: correct the opacity/scale of individual Gaussians.
• Edit in Blender Pro: use the free "3DGS Render" add-on (KIRI Engine) to import, combine and render splats.

5. Phase 4 — Export

The right format depends on where the result is going.

Photogrammetry meshes

Gaussian Splats BeginnerIntermediate

For web embedding, the Khronos Group has been standardizing two glTF extensions since 2025 (KHR_gaussian_splatting and SPZ compression), so splats are increasingly moving into the glTF ecosystem. Otherwise, browser-based viewers (SuperSplat) or integration into Unity/Unreal are enough to view and share them.

6. The most common pitfalls

Now go deeper

This page is the overview. There's a detailed tutorial for each of the two techniques: