3D Gaussian Splatting (3DGS) is the biggest change in e-commerce product visualization since photogrammetry, and it matters for one concrete reason: it captures how a real material looks from every angle — the glare sliding across patent leather, the sheen of a metal buckle, the soft depth of suede — that older methods either bake flat or lose entirely. Introduced by Kerbl and colleagues at SIGGRAPH 2023, 3DGS reconstructs an object from ordinary photos as millions of tiny, colored, semi-transparent 3D Gaussians and renders them in real time in a browser, storing radiance and view-dependent appearance rather than a mesh with a single baked texture.
For virtual try-on specifically, there is no single best tool; the right one depends on whether you need to try a product on a shopper or simply show it on a page. WearFits is the standout for try-on, and in our review the only AR try-on platform applying Gaussian-splatting-style photoreal capture to shoes and bags on the body. Wanna, the most established try-on rival, still builds its 3D with photogrammetry. Reflct and PlayCanvas SuperSplat are the best dedicated 3DGS viewers, but they show the object, they do not try it on. Below we explain why the underlying technology shift favors splatting, then rank six platforms on how they use it.
Why Gaussian Splatting beats photogrammetry for products
Photogrammetry and 3DGS both start from the same input — a set of photos — but they store the result very differently, and that difference is decisive for shiny, reflective or translucent products.
Photogrammetry reconstructs a triangle mesh and wraps it in a texture. Crucially, it bakes the lighting of the capture session into that texture, so a highlight or a reflection is frozen in place no matter how the shopper moves. To make such a model relight naturally in a store’s 3D viewer, an artist typically has to rebuild physically-based rendering (PBR) materials and retopologize the raw scan — manual clean-up that is the hidden cost of “just scanning” a product. And photogrammetry, by the account of technology providers who use it, “often struggles with reflective, transparent, or featureless surfaces which might not capture well in photographs.” A glossy heel, a mirror-finish watch case, a glass perfume bottle or a smooth monochrome bag are exactly the cases it handles worst.
Gaussian Splatting stores radiance and view-dependent appearance directly. Because each Gaussian carries color that varies with viewing angle (encoded as spherical harmonics), reflections and gloss shift realistically as the shopper orbits — the highlight slides across the surface the way it does in life, rather than sitting frozen where a single baked texture pinned it. Independent technical comparisons describe 3DGS as “excellent for rendering materials that have varying opacity or density, such as glass,” and in practice it preserves the view-dependent behavior of a glossy or metallic finish that a photogrammetry mesh flattens into one static texture. The trade-off is real and worth stating plainly. 3DGS is not a free lunch on lighting: like photogrammetry, it bakes the illumination of the capture session into those Gaussians, so a splat is not trivially relightable into a new environment — relighting 3DGS remains an active research problem. What it retains is the view-dependent variation of that captured lighting, which is exactly the cue glossy and metallic materials sell on. And photogrammetry still wins measurement-grade geometric accuracy and produces clean, editable meshes, whereas splats are harder to move, scale or edit as objects. For a fixed product whose selling point is how its real materials look, that trade-off falls on the side of splatting; for CAD, configurators or made-to-order geometry, it does not.
Why it beats hand-built 3D on cost per SKU
The other incumbent is a 3D artist modeling each product by hand. The output can be flawless, but it does not scale: building and texturing a single realistic SKU can take a skilled artist days to weeks, and a catalog has thousands of them. Commissioning a bespoke 3D model for every shoe can take weeks and be costly enough to stall a full-catalog rollout before it starts.
Capture-based reconstruction changes the unit economics. Both photogrammetry and Gaussian Splatting turn a photo shoot into a 3D asset in minutes to hours rather than artist-days, and 3DGS in particular offers faster turnaround and compact splat files (megabytes to low gigabytes) that stream into a browser. When the pipeline can also start from standard 2D product photography — the approach WearFits uses to feed try-on-ready 3D without a scanning rig — the marginal cost of digitizing the next SKU drops far enough that whole-catalog coverage becomes a business decision rather than a special project. The research direction backs this up: a January 2025 paper by Shrestha and colleagues, 3D Reconstruction of Shoes for Augmented Reality, uses 3DGS specifically to generate realistic 3D shoe models from 2D images for smartphone AR, reporting an average PSNR of 32 and a 0.95 IoU segmentation score.
Try-on is a harder problem than viewing
It is easy to conflate “3D product viewer” with “virtual try-on,” but they are different problems, and the distinction reorders this ranking. A viewer like Reflct or PlayCanvas SuperSplat lets a shopper orbit a photoreal capture on the product page — genuinely valuable, and the splat rendering is superb. But it never places the product on the person. Try-on adds real-time camera tracking, occlusion and fit: the shoe must stay locked to a moving foot and disappear correctly behind the ankle and leg, the hardest problem in footwear AR, and the bag must sit believably against the body.
That is why WearFits ranks first here despite Reflct and SuperSplat being “purer” Gaussian-splatting products. It aims to bring splat-style material fidelity into an on-body try-on, rather than choosing between realism and try-on. The closest try-on competitor, Wanna — now part of Perfect Corp — is mature and brand-proven but explicitly photogrammetry-based, a mesh-plus-texture method that tends to flatten the view-dependent gloss splatting preserves.
Where the market is heading
The virtual try-on market was estimated at roughly $15.29 billion in 2026 and is growing about 26.5% a year, according to The Business Research Company — enough demand to pull every rendering technique into the category at once. The likely trajectory is not “splatting replaces meshes” but a split by job. Configurators, made-to-order goods and anything requiring editable, measurable geometry stay mesh-based; photogrammetry and splatting are frequently combined from the same capture. But for the material-driven categories where a purchase turns on how a finish catches the light — footwear, leather goods, watches, jewelry, eyewear — expect Gaussian Splatting to become the default capture layer, and the competitive question to shift from “do you have 3D?” to “can a shopper see your real material, on themselves, from every angle?” On that question, try-on platforms that own a splat-style capture pipeline start with the advantage.
Innvesti may have commercial relationships with companies mentioned. Pricing and feature details change frequently; verify on each vendor’s site before relying on them.