Guide · Configurator technology

Pixel streaming or WebGL?

Pixel streaming renders on a cloud GPU and sends interactive video to the browser; WebGL renders on the visitor’s device. The first gives maximum quality on any machine and costs one GPU per connected user; the second has no per-session cost and demands optimisation. Neither is better in the abstract.

Mori Maison team · Reading 7 min · Updated: November 2026

Where the render happens

The difference is physical: in WebGL the graphics processor is in the visitor’s phone or laptop; in pixel streaming it is on a server. Everything else — quality, cost, latency, scale — follows from that decision.

In WebGL the browser downloads geometry, textures and code, and draws the scene with the device GPU. The heavier the model, the longer it takes to open and the worse it runs on an old phone. In exchange there are no servers in between: a thousand simultaneous visitors cost the same as one.

In pixel streaming the server runs the engine with the full scene, renders every frame and encodes it as video. The browser only receives the video and sends back mouse or touch events. The device needs no GPU, but every session occupies a cloud GPU for as long as it lasts.

How it feels to use

Well implemented, the visitor cannot tell a video from a local render: they turn the piece and it responds. What gives pixel streaming away is latency on poor networks and image quality in motion, especially on large screens with fine detail.

Latency has three parts: the trip to the server, render and encode, and the trip back with the frame. On a good connection it is barely noticeable; on a congested mobile link the drag arrives late. That is why pixel streaming is tested on the real network where it will be used, not only in the office.

Compression shows too. Fast motion with fine detail can produce artefacts on edges, and small interface text blurs if compressed hard. A common fix is to draw the interface in HTML above the video and leave the engine for the 3D scene. Buttons and spec sheets then stay sharp.

The two options side by side

WebGL wins on cost per user, start-up and tolerance of poor networks. Pixel streaming wins on the quality ceiling and on freedom with materials. The decision follows the product, the audience and peak concurrent sessions, not the most spectacular demo.

The cost row is the one most often forgotten in a technical comparison. A configurator with thousands of daily visits can run on WebGL with no variable cost and ruin the budget on pixel streaming, because every visitor occupies dedicated hardware.

WebGL and pixel streaming in a product configurator
CriterionWebGL on devicePixel streaming
Where it rendersOn the phone or laptopOn a cloud GPU
Maximum qualityHigh, with weight limitsEngine-level, scene uncut
Initial weightTarget under 5 MB per productAlmost none: video arrives
MobileTarget 60 fps on mid-rangeDepends on network, not GPU
Cost per userZeroOne GPU per connected session
Start-upDownload and compileConnect to the server
Works on an unstable connectionYes, once loadedNo
ScaleHigh: the device does the workLimited by available GPUs

When pixel streaming makes sense

When the material demands a game engine and the context can take it: automotive, marine, large machinery or a trade-fair screen. Also when the model has reflections, glass, complex lighting or millions of polygons that will not fit on a phone.

The fair case deserves a note. A large screen with a GPU next to it is not the same as a link opened on a distributor’s phone: if the goal is the fair, streaming can be the best option; if the buyer configures from home, probably not.

  • Automotive and marine: paints, chrome, upholstery and studio light.
  • Trade fairs and physical showrooms with a large screen and controlled connection.
  • Products with materials WebGL cannot render fairly.
  • Scenes with many objects, lights or effects that cannot be cut.
  • Guided demos run by a sales rep, with few simultaneous sessions.
  • When the visitor’s device is not a variable you can control.

When WebGL is the answer

In most product configurators: furniture, kitchens, lighting, bathroom, joinery. The product reads well with good PBR materials and optimised geometry, and the visitor arrives from a link without waiting for a server.

WebGL has three advantages that matter in sales: it opens fast, it costs nothing per session and it works anywhere with a normal connection. For a manufacturer who wants a distributor to configure on a phone during a visit, that weighs more than a perfect reflection on chrome.

The limits are known: fewer dynamic lights, approximate reflections and a budget of polygons and textures. It is offset with environment maps, well-calibrated materials and baked lighting. A careful WebGL configurator looks better than a badly compressed streaming one.

The cost per user, explained without rates

In pixel streaming every connected visitor occupies a GPU for the length of the session. Cost grows with concurrency and with server hours. In WebGL that cost does not exist: the hardware belongs to the visitor. We give no figures because they depend on the provider and the usage.

You also have to decide what happens when no GPU is free: queue the visitor, show a WebGL version or images while they wait, or pay for reserved capacity. None is free, and the choice should be made before launch, not during the first campaign.

Start-up counts as well. A GPU server is not instantly available if it is switched off to save money, and booting can take a while. A fair with fixed hours can be planned; unpredictable traffic needs a capacity strategy and a waiting experience that does not scare people off.

Network, latency and accessibility

Pixel streaming depends on the visitor’s network. On fibre it is fine; on patchy mobile coverage it can fail exactly when a rep is showing the product. WebGL, once loaded, keeps working even if the connection worsens.

Accessibility changes too. In WebGL the scene can be paired with accessible HTML: text, sheets, buttons with visible focus and alternatives without 3D. In pixel streaming the scene is a video; the interface has to be built separately in HTML so it can be navigated with a keyboard and read by a screen reader.

And there is the fallback. Either way, an image or a prerecorded video is worth having for anyone who cannot load the scene. A configurator that shows a photo and a contact button still sells something; one that shows an error does not.

All three technologies can coexist

Yes. One project can use pre-rendered images for the catalogue, WebGL for the configurator and pixel streaming for the hero product or the trade-fair screen.

The sensible mix is usually this: images for the product page and campaigns, WebGL for everyday configuration and pixel streaming for a guided demo or a stand. Variable cost is concentrated where it adds value and the rest of the catalogue does not depend on servers.

The complexity is maintaining three outputs from the same model. If geometry and materials are organised, they are reused. If each technology has its own hand-made version, the catalogue falls out of sync in the first season.

How to decide in five questions

Ask how many simultaneous sessions you expect, which device will be used, what materials the product has, how long the visitor will wait and what happens if the network fails. With those answers the technology chooses itself.

  • How many visitors can coincide on the worst day of the year?
  • Is it used mostly on mobile, on desktop or on a trade-fair screen?
  • Are there reflections, glass or lights WebGL cannot approximate?
  • Does the visitor arrive through a link or with a sales rep?
  • What experience do you want when the connection is poor?

Frequently asked questions

What is pixel streaming?

Rendering the 3D scene on a cloud GPU and sending the result to the browser as interactive video. The device does not load the model: it receives the image and returns mouse or touch events.

Can you tell the difference from WebGL?

On a well-optimised product, less than you would think. The difference appears in complex materials, lighting and object count, and in start-up: WebGL loads the model; streaming connects to a server.

What does pixel streaming cost?

We give no rate because it depends on the provider and the usage. The deciding factor is that every connected visitor occupies a GPU for the session, and cost grows with concurrency.

Does it work on mobile?

Yes, as long as the network holds. The device needs no powerful GPU but does need a stable connection. On patchy mobile coverage it can fail, so an image or WebGL fallback is worth having.

Can we start with WebGL and move to streaming later?

Yes, if the model and the rules are separated from the visual layer. Geometry and materials are reused; what changes is the engine that shows them.

Is pixel streaming better for SEO?

Not directly. The indexable content is the HTML around the scene: text, sheets, questions and links. An interactive video provides no text on its own, so the page needs real content besides the 3D.

Which technology do you use by default?

WebGL, because it solves most product configurators and has no per-session cost. Pixel streaming is reserved for automotive, marine, large machinery or trade-fair screens, where engine quality justifies a dedicated GPU.

Sources and references

Related service

We choose the technology by use, not by fashion

We design the configurator with the technology your product and audience can take: images, WebGL or pixel streaming. With the performance target in writing and the code in your hands.