AI rendering terms, defined clearly

Creating or revising an architectural image with a generative image model. The input can be words, a sketch, or a captured 3D view.

It supports visual communication. It is not construction verification.

A captured view from a CAD or 3D application. It carries the visible geometry, camera, framing, overlays, and image proportions.

The CAD Scene mode that starts from a written scene description. It does not require a source image.

The CAD Scene mode that restyles a source image while asking the model to retain its camera, angle, framing, proportions, and major geometry.

Enhance aims to preserve the captured view. It does not rebuild a live 3D model.

The CAD Scene mode that changes marked regions of one source image. One prompt references every unique Region name.

Generative models can still introduce small differences outside a mark.

The image being restyled in Enhance or changed in Edit. It is selected in the mode panel and sent as a direct render input.

A source image is not a Project reference image or Composer reference image.

One optional image in Project settings that can generate the five Composition fields. The image itself is not sent with later renders.

It seeds persistent text. It is not a render source.

An image attached to one Composer message as visual guidance for style, materials, content, or mood.

It guides one send. It does not replace an Enhance or Edit source image.

A stable, editable reference for one marked Edit area. Names such as Region A connect each mark to one instruction in the Edit prompt.

A Region name is control metadata, not requested text inside the image.

The named container for one design effort. It holds Project settings, studies, and renders.

The persistent render context shared by every study in a Project: Composition, the Project reference image, and Project render defaults.

The five project-wide text fields combined: Style, Scene, Lighting, Materials, and Entourage.

Composition is the complete set. Scene is one field inside it.

The Composition field for the overall visual language, such as era, movement, regional character, palette, and finish.

The Style field is one field. A Preset contains all five fields.

The Composition field for reusable environmental context, such as climate, landscape, urban character, and broader atmosphere.

It does not set a camera, viewpoint, room program, or building geometry.

The Composition field for time of day, direction, shadows, weather, atmosphere, and visible artificial light.

The Composition field for important surfaces and finishes, including texture, color, and reflectivity.

The Composition field for people, planting, vehicles, signage, furniture, and activity around the architecture.

An ordered thread of messages and renders inside one Project. Every study shares the same Project settings.

A Study keeps one view or option separate without creating another Project.

A named, reusable bundle of all five Composition fields. A saved Preset can keep immutable versions.

A Preset is not only a rendering style or the single Style field.

A read-only five-field Preset supplied by CAD Scene. Applying one copies its fields into a Project as a starting point.

A recorded checkpoint for one Composition field in one Project.

A Field version records one field. A Preset version records all five.

An immutable snapshot of all five Composition fields under one saved Preset.

Built-in Presets are read-only and have no Preset version history.

One generated image and its stored record, including the prompt, model, settings, credit cost, and Composition snapshot.

The durable server-side generation state for one output image. Parallel outputs have independent jobs.

The model and its supported native choices, including Quality, Resolution, and Create Aspect where available.

Available controls depend on the selected model and mode.

The proportional relationship between image width and height. Create can expose model-supported ratios. Enhance and Edit follow the source proportions.

A model-supported requested output size such as 1K, 2K, or 4K. The measured finished dimensions can differ from the request.

A provider-native generation setting exposed only by models that support it. It is separate from a reviewer judgment about image quality.

Parallel images requested by one Composer send. They share the prompt and settings selection but have separate jobs and charges.

Variants are chosen before Send. Variations start later from a finished Render.

A later configured send started from one finished Render. It repeats the original inputs while allowing supported model, setting, and count choices.

Variations create new renders. They do not duplicate an existing file.

The read-only record of a Render’s stored model, validated settings, cost, Composition snapshot, references, requested size, and output size.

A resize that increases output dimensions. CAD Scene’s free Lanczos 4K upscale does not run another generative model.

Upscaling changes dimensions. It does not add a new design pass.

The unit CAD Scene uses to charge generation actions. Render cost varies by model and supported native settings.

One credit does not equal one Render.

The return of the charged credits for a failed or successfully cancelled Render job. Sibling jobs settle independently.

A later Retry or Variation is a new charged submission.

An image used to discuss design intent, atmosphere, materials, or options with a client.

It can support a design conversation. It is not construction evidence or a verified record of dimensions.

How closely a generated result keeps the visible massing, openings, edges, proportions, camera, and framing of its source image.

A preservation instruction is not a measured guarantee. Compare the result with the source.