Gaussian Splat vs. 3D Model: Which Belongs on Your Product Page
A category-by-category breakdown of when a Gaussian splat beats a 3D model on a Shopify or DTC product page, when the model still wins, and when to ship both.
If your premium DTC product page already has a 3D viewer and the conversion lift has not shown up, the question is rarely whether to add 3D. It is which kind of 3D matches the category. A Gaussian splat and a 3D model both let a buyer rotate a product on a phone, but they sell different things and they cost different things to make. Picking the wrong one is why most 3D experiments stall.
When does a Gaussian splat belong on a product page, and when does a 3D model?
The category decides. A Gaussian splat wins on materials, lighting fidelity, and capture speed; it captures the real product as a cloud of small, colored, semi-transparent shapes from a few photos. A 3D model wins on geometry, animation, and configuration; it is a sculpted mesh with hand-tuned materials. For jewelry, leather goods, ceramics, audio housings, and other products where surface tells the story, ship the splat. For furniture, lighting, sport equipment, and products with crisp edges or moving parts, ship the model. For a premium store with both kinds of products, ship both and let the page pick.
The technique behind splats was published as 3D Gaussian Splatting for Real-Time Radiance Field Rendering at SIGGRAPH 2023 by Kerbl, Kopanas, Leimkühler, and Drettakis at INRIA. The practical result for ecommerce is a 1 to 4 MB file that ships from 2 to 8 product photos.
How they differ in capture
A splat captures what is already there. A merchandiser walks around the product with a phone, takes 2 to 8 photos at spaced angles, runs them through a splat pipeline, and ends up with a buyer-ready file in minutes.
A 3D model is built. A 3D artist sculpts a mesh in Blender or similar, paints PBR materials, sets up lighting, and exports a GLB or USDZ. A premium product takes a half day of artist time, longer for complex geometry. For a 200-SKU catalog, that is the math that decides whether you ship 3D at all.
The cost split shows up in iteration too. If the lighting in your photo studio shifts, a splat can be recaptured in 10 minutes. A 3D model has to be re-sculpted or have its materials retuned by the artist who made it.
How they differ on mobile
Mobile is where premium DTC bleeds. Smartphones drive about 45% of US web traffic per Shopify, and Google's Largest Contentful Paint target is 2.5 seconds for a "good" rating at the 75th percentile. The buyer gets less than 3 seconds before they decide to scroll or leave.
A compressed splat for a single product is 1 to 4 MB. A photoreal GLB with PBR materials is 8 to 15 MB. With Draco mesh compression, a GLB can drop to 1 to 5 MB, but the texture maps that carry the material detail stay heavy unless you cut their resolution. The splat usually wins the size race, especially on a phone over LTE.
The size win matters less if the page does not lazy-load. A 3 MB splat that loads on first paint is just as bad as a 12 MB GLB. Both should fade in only when the buyer scrolls near the rotation block.
How they differ on materials and lighting
This is where splats earn their place. A splat captures the real reflection on brushed gold, the real grain in walnut leather, the real translucence in a ceramic glaze. A 3D model with PBR materials can get close, but it is a guess, not a measurement.
For categories where the buyer is paying for the surface, the splat reads more honest:
- Jewelry. Gemstone refraction and gold finish tone change a $400 ring buyer's mind. A PBR material can fake the reflection; a splat just shows it.
- Leather goods. Stitch detail, grain, and edge paint matter at premium price points. A splat captures the imperfections that signal real leather. A model smooths them out.
- Ceramics. Glaze depth and rim shape sell premium kitchenware. A splat reads the way the kiln finished the piece.
- Audio. Speaker grilles, brushed enclosures, and finish matte vs. gloss decide between $500 and $1,500 buyers. A splat captures all three without an artist tuning materials.
If the buyer is paying for what the surface looks like, ship the splat.
How they differ on geometry and configuration
This is where 3D models earn their place. A splat is photoreal but soft at sharp edges. Stack a few hundred thousand fuzzy dots and you get a beautiful approximation of a curved leather surface; you also get a slightly mushy frame on a chair leg.
For categories where the buyer cares about geometry, a clean GLB still wins:
- Furniture. Buyers measure. A sofa's frame, a chair's joinery, a desk's edge profile have to read sharp. A splat smooths the corner where a model holds the line.
- Lighting. Pendant lamps, sconces, and fixtures sell on silhouette. The clean line of a brass arm reads better as a sculpted mesh.
- Sport equipment. Bicycles, skis, and racquets are geometry products. The frame, the camber, the head shape all need the precision a model gives.
- Configurators. A buyer swapping fabric on a sofa or changing the dial color on a watch needs separable parts. A splat is one frozen capture.
If the buyer is paying for the shape, ship the model.
How they differ on AR and animation
AR placement still favors models. iOS Quick Look ships USDZ. Android Scene Viewer ships GLB. Native AR support for Gaussian splats is improving on both platforms, but the path is uneven and depends on browser support that does not match a buyer's everyday Safari or Chrome session yet.
Animation is the same story. A splat does not animate parts. A 3D model can show a watch crown winding, a kettle pouring, or a car door opening. For categories where the motion is part of the sale, the model is the only choice today.
When to ship both
Most premium DTC catalogs are not all-jewelry or all-furniture. They have hero products that sell on surface and supporting products that sell on shape. The right setup ships both kinds of capture and lets the product page decide which one shows.
A practical pattern that works:
- Splat for the in-page rotation. Loads fast, looks honest, lifts time on page.
- Small GLB for the AR button. Fires only when the buyer taps "see in your space."
- Hero photo for the first paint. Static, brand-correct, fastest to render.
The buyer experiences one product. The page picks the right capture for the question they are asking.
Where to start if you have not shipped 3D yet
If you have a 3D viewer app on Shopify and the conversion lift has not shown up, the issue is rarely the technology. The honest comparison of Arty, LEVAR, Cappasity, and Magic 360 walks through which app fits which category constraint.
If you have not shipped 3D at all, start with a single product. Pick the one where buyers ask the most pre-purchase questions. If that product sells on surface, capture a splat. If it sells on shape, build a model. Put it under the hero with a single proof block that updates when the buyer rotates.
The founder's guide to Gaussian splatting for ecommerce covers the splat capture pipeline end to end. The choice between splat and model is not a technology debate. It is a category question.