Render what you can build: product imagery as an honesty test.

Somewhere right now, a launch render is being polished: seams deleted, vents removed, the wall thickness that physics demands quietly halved because it looks sleeker. Increasingly the image is generated outright, with no CAD behind it at all. Either way it will do its job (clicks, preorders, applause), and then manufacturing will spend a year failing to build it, or customers will unbox something visibly poorer than the promise. The gap has a cost, and the brand pays it with interest.

The vaporware tax

Renders that outrun reality tax a company twice. Internally, they set expectations engineering must then negotiate down, one disappointing review at a time. Externally, they train customers to discount everything you publish: the worst possible asset for a technical brand, whose entire value is being believed. In categories where buyers are engineers, an impossible fillet or a fantasy thinness is spotted instantly. The audience you most need to convince is the one you can't fool.

Built from production CAD

The alternative is structural, not moral: build launch imagery from production CAD, in the same pipeline that serves engineering. Parting lines, real radii, honest materials, rendered beautifully. Because our visualization practice sits next to engineering, the model in the render farm is the model the factory receives, the same model can drive the interactive 3D view on the product page, and assets stay current as the product evolves instead of drifting into fiction.

Where accuracy pays

Accuracy compounds in unglamorous ways: the same asset library serves the keynote and the datasheet without contradiction; the exploded view doubles as service documentation; the how-it-works film is true enough for training. One source of visual truth, many outputs: cheaper than maintaining a beautiful lie in parallel with reality.

The agent-native lens: pictures that agree with the numbers

When the buyer sends an assistant. Assistants that research products on someone's behalf can read images as well as text, and they can set the two side by side. A render that shows a port the specification table doesn't list, or a profile thinner than the published dimensions allow, is a contradiction a machine may surface where a hurried person would scroll past. Keep the pictures and the numbers coming from one source: images rendered from the released CAD, dimensions and materials published as schema.org product data from that same release, and anything conceptual labelled as illustrative wherever it appears. An assistant has no taste to win over. It has consistency to check.

When the business runs on agents. An AI-native visual pipeline can turn out variants far faster than a team working by hand: every colourway, region and angle, re-rendered from the CAD scene. That pace is only safe with checks. Agents can tag each asset with the CAD revision and finish specification it came from, flag every image built on a superseded revision when engineering releases a change, and hold back any colourway that isn't on the approved finish list. Two people sign before anything is published: the art director for how it looks, the engineer for whether it's true. And no image, rendered or generated, shows a feature that hasn't been released.

The refusal

Sometimes a client asks for the shot that can't be true: the impossible thinness, the cable that doesn't exist, the mechanism doing what it can't. Our answer is the same every time: we'll show you what it will actually look like, lit and composed to its best honest angle, and we'll make that image better than the lie. Generated imagery has its place early, for exploring a direction before the CAD exists; it doesn't get to stand in for the product at launch. If a product needs fiction to be desirable, the problem isn't the render. Fix the product. That's the part the rest of the studio is for.