The 3D viewport
A toggle in the viewport's top bar lifts layers with their own 3D switch off the flat page into a shallow depth stage, for parallax and camera moves, not a 3D modeller.
The mode-toggle group in the viewport's top bar has three cells: FX, Camera View and 3D. Turning on the third one drops the whole scene into a shallow 3D stage. Your existing artwork does not change, only how you are looking at and arranging it does. Switching it on for the first time seeds the camera's Camera Depth to 1 if it was still at 0, so multiplane parallax has something to show straight away.
What makes a layer 3D
A layer only leaves the flat page once you turn on its own 3D switch (a per-row dot on its Timeline row, or 3D Layer from its right-click menu), and the switch propagates to that layer's whole subtree. Camera and Bone rows do not get one.
A 3D layer renders at its Position Z depth and can be tilted with Rotation X and Rotation Y, both independent keyframe channels from Rotation, the ordinary 2D roll. Push Position Z back and the layer parallaxes less as the camera pans; tilt it on X or Y and it reads as a card leaning in space rather than a flat cut-out.
3D layers stack by depth, not by their row in the layer list: a layer far behind the others stays behind them even if it sits higher in the list, and the 2D view and the 3D view always agree about the same scene. A 2D layer still works as a divider between 3D layers above and below it, and a layer's Order setting still hops it in front of or behind its neighbour, now counted by depth.
A tilted layer is drawn at its own size on screen rather than as a full frame, so scenes with several tilted layers stay quick, even with onion skin on. Effects and depth of field on a tilted layer are never cut off at a straight edge.
Two ways to look
With Camera View off, you get a free view of the stage: an orthographic camera of its own, independent from the scene's camera, that you orbit by left-dragging empty space. A small axis indicator sits in the bottom-left corner: click its red, green or blue ball to snap straight onto that axis.
Turn Camera View on and you are looking through the scene's actual camera instead, in full perspective, framed to the canvas, and orbiting is locked, because this is exactly what will render.
Moving a 3D layer
Select a top-level 3D layer and a gizmo appears, tilted to match the layer's own rotation. The Move tool gives move arrows only: X and Y in the layer's own plane, Z along its depth. Switch to the Transform tool and the same gizmo adds rotation rings, for spinning the layer about its own X, Y or Z axis, and scale handles. Every drag auto-keys on release.
What this is not
There is no 3D modelling here: no mesh import, no OBJ or glTF, no geometry to sculpt. A 3D layer is still flat, 2D artwork; the 3D viewport only decides where in depth it sits and how it is tilted. The whole pipeline underneath stays a 2D renderer. This is a shared projection bridge for staging layers and the camera in depth, not a 3D engine.