3D streaming pricing works very differently from the way most product configurators are priced. With pixel streaming, you pay for server GPU time for every viewer. With optimized WebGL, the shopper's own device does the rendering, and you mainly pay for the platform and bandwidth.
For one showpiece, either can work. For a catalog of hundreds of products and thousands of daily visitors, the pricing model decides a lot. This post explains the three main ways to deliver 3D on the web, how each is priced, and where each one fits.
The three ways to deliver 3D on the web
1. Pixel streaming
The 3D scene runs on a GPU server in the cloud. The server renders every frame and streams it to the browser as video. The shopper's clicks go back to the server.
Unreal Engine's Pixel Streaming is the best-known version. The engine renders on the host machine and encodes the output into a media stream for the browser.
Strength: very high visual quality, because the server can be far more powerful than a phone.
Cost driver: each active viewer needs GPU time on a server. More viewers, more minutes, more cost.
2. Spatial or asset streaming (e.g. Miris)
A newer approach streams the 3D data itself, adaptively, the way video streaming adapts to your connection. One company building this is Miris. Its docs say you upload an asset, Miris conditions it for delivery, and any browser can stream it without downloads or plugins (Miris docs). Miris says geometry, materials and lighting stream based on what each device can handle, with no server-side rendering per viewer.
Miris opened a public beta in March 2026 (Radiance Fields), and Architosh covered the launch. Its beta page mentions OpenUSD uploads up to 10 GB (Miris beta).
Strength: high fidelity without per-viewer GPU servers, and no manual optimization step.
Cost driver: Miris says you pay for what you stream. It does not publish a price list at the time of writing (checked October 2026).
3. Optimized WebGL (download and render locally)
The 3D model is compressed and sent to the browser, and the shopper's device renders it with WebGL. This is how most product viewers and configurators work, including Polymuse.
The model is made small with standard tools:
- ✓Geometry compression with Draco or meshoptimizer.
- ✓Texture compression with KTX 2.0 and Basis Universal, which Khronos says cuts download size and GPU memory.
- ✓Levels of detail (LOD), so a light version shows first.
- ✓Progressive loading, so the shopper can interact before the full-detail model arrives.
Strength: cost doesn't grow with viewer minutes, and the same file works in AR.
Cost driver: the platform plan and bandwidth. The rendering is free, because it happens on the shopper's device.
3D streaming pricing, with real numbers
Here's what published prices look like for pixel streaming. These are vendor list prices at the time of writing, so check before you budget.
| Vendor | Model | Published rate |
|---|---|---|
| Arcware | Pixel streaming for Unreal | €0.15 per minute on Lite, down to €0.10 or less on Core and €0.08 or less on Enterprise |
| Vagon Streams | Pixel streaming, pay per minute | About $0.025 to $0.047 per minute depending on GPU, plus $0.67 per day per region to keep an app available |
| Miris | Spatial streaming | Not published (beta) |
Those per-minute rates look small. Now do the math for a product page.
Say 20,000 shoppers a month open a configurator and spend 3 minutes in it. That's 60,000 streaming minutes.
- ✓At €0.15 per minute: €9,000 a month.
- ✓At €0.08 per minute: €4,800 a month.
- ✓At $0.025 per minute: $1,500 a month.
- ✓At $0.047 per minute: $2,820 a month.
That's streaming only. It doesn't include the platform, the 3D content or the app build. And it grows with every extra visitor and every extra minute. For a showroom kiosk or a car launch, that's fine. For a furniture catalog where most visitors browse briefly, it adds up fast.
Optimized WebGL pricing
With WebGL, you don't pay per viewer minute. You pay for:
- ✓The platform. The viewer or configurator plan.
- ✓Bandwidth. Usually bundled into the platform, or served from a CDN.
- ✓3D asset optimization. Either done by your platform, by tools, or by a 3D artist.
On Polymuse, views are bundled into each plan: 60,000 a year on Visualize, 100,000 on Configure and 250,000 on Modular. Viewer and configurator traffic never uses AI tokens. An automatic model optimiser is included on every plan. See pricing.
Since January 2026, Polymuse compresses models with KTX2 and Draco, generates LODs for progressive loading, uses WebP textures, and lets you set texture resolution and polygon count per model (January 2026 update). In practice, a 30 MB or 90 MB model keeps its quality and becomes usable in about 5 seconds.
The trade-offs for a large catalog
| Pixel streaming | Spatial streaming | Optimized WebGL | |
|---|---|---|---|
| Cost grows with | Viewer minutes and concurrent users | Data streamed (no public price) | Platform plan and bandwidth |
| Visual ceiling | Highest | High | Limited by the device |
| Mobile | Needs a good connection the whole time | Vendor says it adapts to the device | Depends on optimization |
| AR | Harder, the scene is on a server | Check with vendor | Works, the file is on the device |
| Latency | Every click goes to a server and back | Check with vendor | None once loaded |
| Lock-in | App built for one engine | Assets converted to the vendor's format | Open glTF files, if your platform lets you export them |
| Setup effort | Build and host an app | Upload source assets | Optimize models (often automatic) |
Cost per view
This is the big one. Streaming costs grow with traffic. WebGL costs mostly don't. For a catalog with many products and lots of short visits, WebGL usually wins on cost by a wide margin.
Fidelity
Pixel streaming can show film-quality lighting. Most furniture shoppers don't need that on a product page. They need accurate fabric, scale and colour, which a well-built WebGL model gives you. For hero imagery, render stills offline.
Mobile and AR
AR on a phone needs a local 3D file. That's natural for WebGL and awkward for pixel streaming. If AR matters to your customers, keep that in mind.
Lock-in
glTF is an open standard. Your optimized models work in other viewers, in AR and in renderers. A streaming app or a proprietary streamable format is harder to move. Check whether your platform lets you take the files back out. On Polymuse, glTF export is on the Modular plan and above.
When to choose each one
Choose pixel streaming when:
- ✓You have a single, very heavy scene, like a car or a building.
- ✓Visual quality must be the absolute maximum.
- ✓The audience is small or controlled, like a showroom or an event.
Consider spatial streaming when:
- ✓Your assets are huge (scans, CAD, big scenes) and you don't want to optimize them.
- ✓You're comfortable with a beta product and quote-based pricing.
Choose optimized WebGL when:
- ✓You have a large catalog with many products.
- ✓Traffic is high and visits are short.
- ✓You need AR and open files.
- ✓You want costs that don't jump with every marketing campaign.
The real problem is usually optimization
Most brands look at streaming because their current configurator is slow. That's almost always an optimization problem, not a delivery problem. Huge source files, uncompressed 4K textures and loading every option upfront will make any viewer crawl.
We cover the causes and fixes in Why 3D Product Configurators Load Slowly (and How to Fix It). For the visual side of the trade-off, see Product Visualization. For the full budget picture, see How Much Does a 3D Product Configurator Cost?.
FAQ
How does 3D streaming pricing work? Pixel streaming services usually charge per streaming minute, sometimes with a monthly plan on top. Cost grows with the number of viewers and how long they stay. Spatial streaming vendors like Miris charge for what you stream, but Miris doesn't publish prices yet.
What is Miris pricing? Miris does not publish a price list at the time of writing (October 2026). It's in public beta, and its site says you pay for what you stream, without GPU servers. Contact Miris for a quote.
How much does pixel streaming cost? Published rates we found range from about $0.025 per minute (Vagon Streams, entry GPU) to €0.15 per minute (Arcware Lite). At 60,000 viewer minutes a month, that's roughly $1,500 to €9,000 a month for streaming alone.
What does 3D asset optimization cost? It depends on who does it. Many platforms include it. Polymuse includes an automatic optimiser on every plan. If a 3D artist does it by hand, it's part of the model cost, often within the $100 to $500 per model range.
Is optimized WebGL good enough for furniture? For most furniture product pages, yes. With compression, LOD and progressive loading, large models load in seconds and show accurate fabric and scale. Keep offline rendering for hero images.
Need fast 3D across a big catalog?
See how your heaviest model loads with Polymuse.