Gaussian Splatting vs Photogrammetry: What's the Difference and Which Should You Use?

Two technologies capture the real world in 3D, but they solve fundamentally different problems. One gives you geometry you can measure, print, and edit. The other gives you visuals so realistic they're almost indistinguishable from a photograph. Choosing the wrong one wastes time. Here's how to pick the right one.

What Photogrammetry Actually Produces

Photogrammetry builds 3D models through triangulation — a mathematically grounded process tied to the physical geometry of whatever you're scanning. You take overlapping photos of an object or scene, and the software calculates the precise position of thousands of surface points by comparing where they appear across multiple images.

The output is an explicit mesh: vertices, edges, faces, UV-mapped textures. A real 3D model with measurable dimensions, exportable to standard formats like OBJ, STL, GLB, FBX, and USDZ.

This matters because the mesh carries metric accuracy. You can measure distances on it. You can 3D print it. You can drop it into a game engine, a CAD tool, or an AR viewer. The geometry corresponds to the actual surface of the object you scanned.

In Replica, the entire pipeline runs locally on your Mac — from photo import through alignment, dense reconstruction, meshing, and export in multiple formats. No cloud upload, no waiting for remote servers.

What Gaussian Splatting Actually Produces

3D Gaussian Splatting (3DGS) takes a different approach. Instead of reconstructing geometry, it learns a scene's appearance. The algorithm places thousands of small, semi-transparent blobs — Gaussians — in 3D space and optimizes their position, size, color, and opacity until the rendered view matches the original photos from every angle.

The result looks stunning. 3DGS can render photorealistic views at 90+ frames per second, handling things that trip up traditional methods: hair, foliage, reflective surfaces, translucent materials. Visually, it often outperforms photogrammetry.

But there's a catch. Those Gaussians aren't geometry in any traditional sense. There are no vertices to edit, no faces to slice, no surface to measure. The representation is optimized for how things look, not for what things are.

Where Each Technology Excels

Photogrammetry wins when you need real geometry

• 3D printing. A photogrammetric mesh with object masking enabled comes out watertight and ready for slicing. Gaussian splats can't go directly to a printer — you'd need an intermediate conversion step, and the resulting mesh is typically less clean.
• Measurement and engineering. If you need to measure distances, fit parts, or integrate with CAD/BIM/GIS workflows, photogrammetry gives you metrically valid data. Gaussian splatting optimizes for visual fidelity, not dimensional accuracy.
• Editing and post-processing. Standard mesh tools (Blender, Maya, ZBrush) work directly with photogrammetric output. You can decimate, retopologize, UV-unwrap, or boolean-cut the model. With 3DGS, your editing options are limited to specialized splat editors.
• AR and e-commerce. Product models for AR Quick Look, web 3D viewers, or e-commerce platforms need proper meshes with clean topology and optimized textures. That's photogrammetry's natural output.

Gaussian splatting wins when you need visual immersion

• Virtual tours and real estate. Capturing an apartment or venue for an interactive walkthrough? 3DGS produces photorealistic results faster than photogrammetry, and the viewer experience is smoother. Zillow and other platforms have started adopting it for property tours.
• Virtual production and previs. Film and game studios use splats for set scouting, lighting tests, and LED wall backgrounds. The 2025 Superman production was among the first major films to use dynamic Gaussian splatting in its pipeline.
• Heritage and documentation (visual). When the goal is to preserve the appearance of a space — a museum gallery, a historical streetscape — 3DGS captures lighting, reflections, and atmospheric detail that mesh-based methods flatten.
• VR experiences. The real-time rendering performance of 3DGS makes it well-suited for VR, where maintaining high frame rates is critical. Unreal Engine added 3DGS support, and VRChat communities have started building photorealistic environments with splats.

The Convergence Question

In 2025, the Khronos Group proposed adding Gaussian Splat support to the glTF format — a signal that 3DGS is moving from research into production tooling. Tools like SuGaR and KIRI Engine's 3DGS-to-Mesh pipeline now bridge the gap, converting splats into editable meshes.

Does that make them interchangeable? Not yet. The converted meshes are improving, but they still don't match the geometric precision of native photogrammetric reconstruction. If your downstream task requires accurate geometry — printing, measurement, engineering — starting with photogrammetry saves you a conversion step and gives you better results.

The two technologies are converging, but today they remain complementary. The best approach depends on what you need the 3D data for, not which one is "better."

A Decision Framework

Ask yourself three questions:

1. Do I need to measure, print, or edit the 3D model? Yes → photogrammetry. You need real geometry with metric accuracy.

2. Is the primary goal a photorealistic visual experience? Yes → Gaussian splatting. It renders faster and handles complex visual materials better.

3. Am I capturing a single object or an entire environment? Single object → photogrammetry tends to be more practical, especially with features like object masking that isolate the subject automatically. Full environment → 3DGS shines for immersive capture.

If you answered "both" to questions 1 and 2, consider running both pipelines on the same photo set. Capture once, process twice — use the photogrammetric mesh for your editable asset, and the splat for your visual presentation.

Where Replica Fits

Replica is a photogrammetry tool. It produces explicit, measurable 3D meshes from photographs — processed entirely on your Mac using Apple Silicon. If your workflow involves 3D printing, product scanning, AR content, or any application that requires a real mesh, that's the tool's sweet spot.

What Replica doesn't do is Gaussian splatting. If your goal is an immersive virtual tour or a photorealistic VR environment, a dedicated 3DGS tool is a better fit for that specific task.

The honest answer is that most practitioners will use both technologies at different points. The question isn't which one replaces the other — it's which one you reach for first given the job at hand.

Getting Started

If you're working with objects, products, or anything that needs to become a tangible 3D asset, start with Replica. From quick STL exports to multicolor 3D prints, the pipeline is built for practical output.

If you want to push photos from your phone directly to your Mac for processing, Replica Link turns your Mac into a local photogrammetry server accessible from any device on your network.