Working in 2D and 3D
Flat design and 3D share one space. What you make decides where it starts.
There is no global 2D-or-3D mode. What you make decides where it is created: a mesh, scattered points or a gaussian splat in the 3D scene; a flat shape, text or an imported image as a 2D composite layer.
Both sit in the same space. Orbit the camera and flat work is standing in it. Anything moves between the scene and the layers at any point, with no conversion step.
One space, no mode switch
Circles, rectangles, lines, text and pen paths appear flat and front-on.
Spheres, imported models and scattered points land in the 3D scene.
The two mix freely in the same world.
New flat content behaves as a stacked design layer, which is what posters, titles and UI mockups need. 3D content behaves as a scene object. Neither is declared; the kind of thing you made determines it.
What goes where by default: 3D geometry, points, gaussian splats, volumes, and 3D shapes start in the scene; flat vector shapes & text, images (rasters), and procedural colour/greyscale fields start as composite layers. (Among fields it splits by kind: vector and distance fields are scene; colour and scalar fields are layers.) Promote or demote anything by dragging it in the Layer Manager or with the Layer node (see Layers & compositing). If a node outputs both 2D and 3D, the 3D wins and the whole node joins the scene.
The viewport's orbit gesture, Alt+drag, works on a purely 2D document. Flat artwork already sits in 3D space, and orbiting shows it from the side.
Camera styles
The viewport has two, and switching between them keeps the view stable.
Orthographic is a head-on look with no perspective. Pan and zoom behave like a normal canvas, which is the natural way to work on 2D layouts.
Perspective has depth and vanishing points, for working in the scene.
You can also look through a camera placed in the scene, as you would cut to an active camera. Navigating then drives that camera directly.
How a flat shape is placed in 3D
Each shape, each letter, each row carries its own placement: a position, an orientation and a scale. Tilt a flat logo so it recedes into the distance, stand text up like a billboard, lay a panel flat on the ground. The artwork itself stays flat.
Tilting a flat shape by its orientation does not convert it into 3D geometry. It stays light and sharp, angled in space. Only content pushed into real depth becomes scene geometry.
Whole-object placement is a separate set of controls from those per-row values. Position, Rotation, Scale and Anchor sit on the node that made the content, with Anchor the pivot the other three orbit. All four are wireable pins and not widgets alone, so a driver node can supply a placement, and a single axis can be driven on its own. Moving them, by gizmo or by wire, re-places the object without rebuilding what feeds it, which is why dragging a shape around in 3D stays responsive on a heavy graph.
The per-row orientation value is the @orient attribute, and flat shapes can also carry @analytic (see Analytic shapes). Placement tools and nodes set both, so you rarely edit them by hand, but either can be driven from an Expression.
Vector work in a scene
Flat shapes are resolution-independent. Edges stay clean at any zoom, because the picture is drawn at the size and angle it ends up at rather than being baked into pixels early. That holds when the shape is tilted in perspective or seen through a scene camera.
Vector work and the 3D scene are rendered together into the same image. A flat title can sit in front of a 3D product shot, or behind it, or be lit by the same lights, in one pass.
Stacking flat work into layers
For stacked compositing with blend modes and opacity, promote content into ordered layers. Flat design falls into that stacked behaviour naturally, and a single layer can contain a whole 3D scene of its own.
Layers are an arrangement of the objects you already have, not a separate compositing document. See Layers & compositing for how stacking, blend modes and track mattes work.