AI rendering

Traditional rendering engines simulate light physically: they trace rays from light sources through a 3D scene, bouncing off surfaces with defined materials. That is accurate but slow to set up, since every material, light and object has to be modeled. AI rendering works differently. A model, usually a diffusion model, has learned what buildings, materials and light look like from a large number of images, and generates a picture guided by an input image, a prompt and a style. It does not need a complete 3D scene, which is why a sketch or a plain screenshot is enough to start.

How it shows up in a project

• It is fastest at the start of a project, when many options need to be seen quickly and none are final.
• It works from what you already have: hand sketches, model screenshots, plans, elevations or site photos.
• It is plausible, not measured. Daylight levels, exact material colors and fine details should be checked against the design.
• Results vary between runs, so generating a few and choosing is a normal part of the workflow.
• Many studios use it alongside traditional rendering, with AI for exploration and edits and ray tracing where accuracy is contractual.

In Rendair

Rendair is built around AI rendering: apps such as Sketch to Render and 3D Model to Render generate images, other tools edit, upscale and animate them, and boards collect the results. See render for the general term.