What a Gaussian Splat Is, and Where One Fits in an Association's Plans

An association wants members who will never travel to its historic headquarters to see the founders’ room and its glass cases of artifacts, and the events team wonders whether the same approach could show next year’s meeting venue. Someone suggests a Gaussian splat, and nobody can say what that is. This is a teaching example, not a case study. What the team has to settle is whether a splat beats familiar photo tours at that job, and what the association would have to maintain.

In April 2020, during the pandemic, Associations Now, ASAE’s own publication, reported that “many venues and destinations are relying on virtual planning tools to help get the job done,” with “venue tours and site visits” moving online as 360-degree and live-video tours. One convention and visitors bureau in that 2020 pandemic piece offered a 360-degree virtual reality tour of its convention center.

How a stack of ordinary photos becomes a scene

The Khronos Group, a standards body, defines the technique this way: “Gaussian splatting is a radiance field representation technique that converts multiple 2D images into a photorealistic 3D asset.” Overlapping photos of a room go in, and out comes a 3D scene a viewer can move through on a screen. There is no shell of walls and floors underneath; a 2025 museum study by Kwon and Yu says the method gives “real-time visualization of complex scenes without mesh modeling.”

The scene is instead a crowd of “thousands of 3D Gaussian primitives,” which Virginia Tech University Libraries call “data-rich points.” The name comes from the bell-shaped Gaussian curve of statistics: each point is strongest at its center and thins toward its edges, so the points overlap into what reads as one continuous room. Khronos explains the fitting: “Machine learning optimizes these Gaussian points to match the original input images.” Kwon and Yu add that the points encode “view-dependent radiance,” so a point’s color can change with the viewing angle, the way a polished floor does as someone walks past.

The museum study notes that “360° image-based systems limit user movement to fixed viewpoints, reducing spatial exploration,” while laser scans and photogrammetry models “offer free navigation and greater immersion, but struggle with reflective surfaces and complex lighting.” In an April 2026 presentation, Vincent Rossi, the Smithsonian’s 3D Program Supervisor, put it this way: “Unlike photogrammetry, splats skip retopology, cleanup, and texture baking entirely.”

A one-page fit sheet for a proposed splat

Fill in the third column before anyone books a shoot. The middle column is our reading of the sources, untested at any association.

Question for the project	What the sources report	What to write down
Will remote visitors move around or look from one spot?	In a 2025 study, a free-roaming splat tour beat a 360-degree tour of the same rooms on every presence item.	The route a visitor should be able to take.
Must they read plaques, signs or labels?	Participants in that 2025 study found the 360-degree tour clearer and the splat blurry on artifacts and text.	Each piece of text that has to stay legible.
Are there glass cases, white walls or strong sun?	The 2025 authors report sunlight noise on floors and noisy blank walls; Virginia Tech’s plan hoped splats would render reflections better.	The surfaces most likely to misbehave.
Is this a public view or the lasting record?	The Smithsonian 3D program’s April 2026 view puts public uses in its use-now column and holds splats back as a preservation master.	Which file is the record and which is the display copy.
Who keeps the file usable as formats change?	The same deck warns that no tested migration path exists yet.	A named owner and the formats to keep.
How much effort will capture take?	The 2025 museum team found the splat took more time and resources than 360-degree images.	The staff hours and storage available.

What the museum study measured, and what it did not

Kwon and Yu, of the Korea National University of Heritage, built one university museum interior twice, as a 360-degree photo tour and as a splat scene. They tested both with 30 participants between March 27 and April 4, 2025. On every presence item, meaning the sense of being in the place, the 2025 splat version scored higher than the 360-degree version, and each difference was statistically significant. Only the splat version let visitors roam, so the result cannot be credited to image quality.

The picture actually favored the older format. Participants said “the 360° image-based system offers high visual fidelity and clarity,” and their complaints about the splat centered on “lower image resolution, blurred rendering of artifacts and text.” Several suggested combining the sharp 360-degree images with the splat’s freedom to move. Participants called the splat tour “a realistic alternative to physical museum visits, especially for individuals unable to travel due to physical constraints,” an opinion rather than a measured outcome.

Where cultural institutions are placing splats for now

Rossi’s April 2026 deck gives his program’s current view under a slide titled “Where we land (so far).” Its use-now column lists “Virtual exhibitions and galleries” and “Site documentation and environment capture.” Its hold column reads “Not as preservation master format” and “3DGS not appropriate for most research use-cases” (3DGS is shorthand for the technique), and it says to “Maintain mesh outputs as archival deliverable.”

Virginia Tech University Libraries described a plan, still in progress when published in 2025, to record a remote sculpture trail, partly as splats. Its aims include “a lightweight, web-performant background environment” and the hope that “it will enable people to experience it anywhere without physical accessibility concerns.” The team notes that “even less guidance is available for AR/VR, Gaussian splats, and volumetric capture” than for 3D models, and it plans on “using open formats where possible, metadata that accounts for technical details.”

On February 3, 2026, Khronos announced a release candidate for an extension that “enables storing 3D Gaussian splats in glTF,” a 3D file format, warning that “Without standardization, this rapid evolution could easily lead to fragmentation.” The Open Geospatial Consortium, another standards body, proposed “3D Gaussian Splatting” for the scope of version 2.0 of its 3D Tiles standard, with comments due by July 23, 2026, and we did not find the outcome.

Filling in the sheet for the founders’ room

Back at headquarters, the route row reads: from the entrance to the cases. The plaque row matters more, since founders’ names must stay readable, so the team pairs the splat with sharp still photos of each plaque. The glass cases and a sunny window go in the trouble row. The record row keeps existing photographs as the archive and the splat as a display copy, and the owner row names whoever checks the file each year.

The mistakes that make a splat project disappoint

The first is telling a board that splats feel more real because they look better, when the study’s splat tour also let people move and lost on sharpness. The second is treating the splat as the archive, the role the Smithsonian program’s deck holds it back from. The third is promising members a benefit nobody has shown. No published source we read shows a trade association, professional society or other membership association capturing or publishing a splat, or measures whether members want one. A pilot on one room, judged against the sheet, is the fairer test.

To make a first splat from photos, follow our companion tutorial; once a scene exists, our guide to putting a venue scan on your website covers the next step and where the splat standard stands now.