My updated 3DGS workflow: from multi-camera capture to final 9:16 / 16:9 export

Over the past few weeks, I’ve rebuilt my entire 3D Gaussian Splatting workflow, especially the post-production stage. The biggest improvement came from switching away from Postshot’s 2D PNG workflow in Premiere and instead importing my splats directly into After Effects with a real 3D camera. This gives me far more control over movement, composition, depth, and timing.

Because my After Effects master composition is 3840 × 3840 px, I can easily link it to two Premiere sequences for clean 16:9 and 9:16 exports without rebuilding anything.

Working directly in AE also lets me manipulate the splats in true 3D (via Irrealix), instead of being limited to 2D exports in Premiere.

Below is my complete pipeline: capture, prep, reconstruction, animation, and export.

My Current End-to-End Workflow

1. Capture

XangleCS – Trigger & Transfer

2. Pre-processing

Lightroom – Color, Tone & Denoise

  • Basic color grading

  • AI noise reduction

Topaz Photo AI – Upscale / Enhance

  • I sometimes do a 1x upscale before reconstruction

  • Skipped this step for this dataset

3. Camera Calibration & 3D Reconstruction

RealityScan (or RealityCapture) - Camera Calibration

Postshot – 3DGS Training

  • Train Gaussian Splats (splat3 model)

  • Export .ply file

4. Animation & Rendering

After Effects - 3D Import (Irrealix plugin)

  • Import splats as true 3D objects

  • Add AE’s native 3D camera for precise movement

  • Use various Boris FX

  • Work in a square 3840×3840 master comp for universal output

5. Editing & Delivery

Premiere – Aspect Ratio Versions

  • Import the AE comp twice (Dynamic Link)

  • Sequence #1: 16:9 export

  • Sequence #2: 9:16 export

  • Add sound design (SFX), music

  • Export in ProRes or h.264 depending on platform


Vertical and horizontal versions, both in 4k

With Hugo Balek at the Xangle Studio.

Next
Next

iPhone 17 + genlock? Is it the next big thing for multi-camera installations?