Visualisation covers architecture, engineering, training and monitoring — cases where a 3D model stands for a real thing and decisions are made from it.
The distinguishing requirement
Accuracy is functional. In games, a model that looks right is right. Here, a model that looks right and measures wrong can produce a bad decision — a clearance that is not there, a fit that does not work, an evacuation route that is blocked.
That changes the source. These models come from CAD, survey data or scans, not from approximation.
Architectural visualisation
Buildings before they exist. Requires real dimensions, correct materials, and lighting simulation that reflects actual sun position and glazing. The aesthetic bar is high because the output competes with photography.
Digital twins
A live model of a real asset, connected to sensor data. The 3D part is the interface; the value is in what it displays. Requires a model accurate enough to locate real components, and a pipeline for keeping it current as the real thing changes.
Training and simulation
Accuracy of behaviour matters as much as appearance — the procedure being trained has to match reality. Visual fidelity is often secondary to correctness.
Where generated models help
Context and dressing. The surrounding environment, vegetation, furniture, vehicles and people around an accurate central model do not themselves need to be accurate — they need to be plausible and cheap.
That division is the useful one: measured geometry for what is being decided about, generated geometry for what surrounds it.
Where they do not
Anything load-bearing in the decision. A generated approximation of a real component in a digital twin is a false statement about the real world, and it will eventually be treated as true.