DAZAI CHEN

PostShot: The Easiest Way to Train 3D Gaussian Splatting

A guide to PostShot - a user-friendly desktop app for creating high-quality 3DGS scenes from photos or video.

3D Gaussian Splatting AI PostShot Photogrammetry Learning

What is PostShot?

PostShot is a desktop application designed to make 3D Gaussian Splatting accessible to everyone. Unlike command-line tools that require technical expertise, PostShot provides a clean, intuitive interface for training 3DGS models from your photos or video.

Don’t know what 3DGS is? Read my introduction to 3D Gaussian Splatting first.


Key Features

Smart Image Selection

One of PostShot’s standout features is Select Best - an AI-powered tool that automatically filters your input images:

  • Removes blurry or motion-affected frames
  • Eliminates redundant/duplicate views
  • Keeps only the most useful images for training

In my Nankunshen Temple project, this reduced 2000+ frames down to ~300 optimal images automatically.

Real-time Training Preview

Watch your 3DGS scene come to life as it trains. PostShot shows:

  • Current training step
  • Live 3D preview of the scene
  • Quality metrics
  • Estimated time remaining

Pause & Resume

Training can be paused and resumed at any time. This is incredibly useful for:

  • Checking intermediate results
  • Adjusting parameters mid-training
  • Extending training if quality isn’t sufficient

Simply increase the Stop Step or Max Splat count and continue training.

Multi-format Export

Export your trained models in various formats:

  • .ply - Standard format, works with most viewers
  • .splat - Compressed format for web viewing
  • .pshv - PostShot’s native format (includes Unreal Engine plugin support)

Training Parameters

Here are the key parameters you’ll work with:

ParameterDescriptionRecommendation
Image SizeMax resolution for training1920 for quality, 1080 for speed
Max SplatMaximum number of Gaussians3-6 million for detailed scenes
Stop StepTraining iterations30,000-40,000 for good results
Select BestAuto image filteringAlways enable

My Typical Settings

For a detailed indoor scene like the Nankunshen Temple:

Image Size: 1920
Max Splat: 6,000,000
Stop Step: 40,000
Select Best: Enabled

Training time: ~45 minutes on RTX 4080 Laptop


Workflow Tips

1. Start Small, Then Extend

Begin with conservative settings:

  • Max Splat: 2,000,000
  • Stop Step: 20,000

Review the result. If it looks promising, increase the values and continue training.

2. Quality vs Speed Trade-off

PriorityImage SizeMax SplatStop StepTime
Quick preview10802M15,000~10 min
Balanced19204M30,000~30 min
Maximum quality19206M+40,000+~60 min

3. Input Image Quality Matters

PostShot can only work with what you give it:

  • Avoid motion blur
  • Ensure consistent lighting
  • Capture from multiple angles
  • Overlap between views is important

Unreal Engine Integration

PostShot includes a free Unreal Engine plugin that allows you to:

  1. Import .pshv files directly into UE5
  2. View 3DGS scenes in real-time
  3. Combine with VR for immersive experiences

This is what I used for the VR experience in my Nankunshen project.


Pricing

PostShot offers:

  • Free trial - Limited exports
  • One-time purchase - Full features, no subscription

Check the official website for current pricing.


Alternatives

While PostShot is my go-to for local training, here are some alternatives:

ToolTypeBest For
PostShotDesktopBest UX, UE integration
Luma AICloudQuick results, no GPU needed
NerfstudioOpen sourceCustomization, research
PolycamMobileOn-device capture & processing

Conclusion

PostShot strikes an excellent balance between ease of use and professional results. If you’re serious about 3DGS but don’t want to deal with command-line tools, it’s the best option available.

The combination of smart image selection, real-time preview, and Unreal Engine integration makes it particularly valuable for VR/XR applications.



Sources


Get in Touch

Have questions or want to collaborate? Feel free to reach out!

dazai.studio

Dazai Chen

dazai.studio@gmail.com