Skip to content

3D ​

Making a layer 3D ​

Add a Transform 3D to a layer's chain, or click the layer's 3D switch in the timeline, which adds one just before Output. The layer becomes a card: its whole picture on a flat plane in the comp's 3D space. At its default values a card looks exactly as the 2D layer did.

The 3D switch is a shortcut over the graph. Off bypasses the layer's Transform 3D nodes (their values and keys stay); on brings them back.

Transform 3D moves what comes before it ​

One rule covers every place a Transform 3D can sit: it moves whatever comes before it, in 3D space, and it never flattens anything.

  • Before it is a flat picture: the picture becomes a card.
  • Before it are points, shapes or text: it moves them, and the Draw node after it draws them in depth.
  • Before it is a picture of things already in 3D (after Draw Points on 3D dots, after a Merge of 3D things): it moves those things. A ball of dots turns as a ball, and a camera can still fly through it.

So Transform 3D nodes stack the way Transform 2D nodes do, on either side of a Draw node or a Merge. For a flat picture of something 3D, put it in its own comp and bring that comp in: a nested comp arrives flat.

Points and shapes in 3D ​

A Transform 3D can also take points, shapes or text, wired in like a Transform 2D. It moves them in 3D, and the Draw node after it draws them through the camera with real depth: dots shrink with distance, and each shape crosses cards and the others correctly.

Where it sits decides what turns. Before Clone to Points it turns the grid's layout and each clone stays facing front; after Clone to Points it turns the clones too. The layer's 3D switch turns these Transform 3D nodes off and on as well.

A Transform 3D after the Draw node moves the same points in depth. The one difference is scale: on the points it spreads them and leaves each dot its size, and after the Draw node it grows the dots and strokes too, as scaling a card grows its picture.

Models ​

Drag a glTF (.gltf, .glb), FBX, OBJ, PLY, XYZ or Alembic (.abc) file into the timeline or the Viewer to get a model layer: Model → Transform 3D → Output. The camera draws the model as it was made — its materials, textures, skins, morphs and animation — lit by an environment image (the Model node's Lighting section; empty uses a soft built-in studio). Alembic caches play their animated meshes and particles from disk. A Camera layer can follow a model file's own camera (glTF, FBX or Alembic): pick the Model layer in the Camera's From Model. PLY and XYZ point clouds come in as dots, and a 3D Gaussian splat PLY as splats. The Model node's Points output hands the file to Caddis's point nodes as points with depth — a cloud, splat centers, or a mesh's vertices — for Draw Points, Clone to Points or an emitter. It crosses cards and other 3D layers by depth, takes motion blur, and parents and is parented like any 3D layer. An effect after its Transform 3D works on what the camera sees of it.

A Draw Shape fill or stroke texture (Canvas mode) turns with a 3D shape, in perspective, as it would on a card.

Grid's Count Z stacks its pattern through depth, so Rows, Columns and Count Z make a box of points, its Sphere pattern spreads points evenly over a ball (or through it, with Fill), and Point Scatter's Depth spreads its points through depth.

Wire a Model into Point Scatter and the points land on the model's surface, evenly by area, and ride it as the model animates.

Points on a sphere or a model know which way they face (a normal attribute). Clone to Points and Draw Instances lay each clone or image on that facing (Follow Normal, shown when the points carry one), and a Point Emitter's Outward fires along it.

Connect Points measures in depth: Nearest and Radius link true neighbors in a 3D cloud, and the lines are drawn through the camera. Draw Instances stamps its images through the camera too: sized by distance, sorted by depth, facing the camera unless their points carry a facing.

Depth is the points' z attribute, so in a layer with such a Transform 3D, Point Attributes can set it like any attribute: a Noise field on z pushes a grid into a landscape.

Particles in depth ​

Particles are points, so they come along: give a Point Emitter a source that went through a Transform 3D and its particles are born in depth. Their velocity gains a third number, their speed in depth, which the Point Solver moves them by (Drag and Max Speed slow it too, and forces push in depth — see Fields in 3D). Flocking works in 3D, the Boundary is a box as deep as the comp is wide, and a collider shape placed with a Transform 3D is a wall in its own plane: particles bounce, slide, stick or die where they meet it. Gravity stays on screen; wire a Vec3 value into Forces for gravity in depth. Outward from a single point bursts over a whole sphere, and Point Trail follows them through 3D.

Fields in 3D ​

Every field is a 3D field. Points and particles with depth read fields at their depth: Noise and Random vary through depth, a Noise or Curl force pushes in depth as well as across the screen, Radial Force pulls toward its center in 3D, and Field Math, Remap and Field Gradient carry depth through. Fields that are flat by nature (images, gradients, shapes, patterns) are the same at every depth. Looping noise is the same at every depth too.

A Vec2 or Vec3 value wired into a field input is the same push everywhere: a Constant (0, 0, 1) into a Point Solver's Forces is gravity in depth.

Effects before and after Transform 3D ​

The order in the graph decides what an effect works on. Before a flat picture's first Transform 3D, it changes the picture on the card and turns with it. After it, it works on what the camera sees of that layer: a Glow after the Transform 3D glows evenly around the turned card, while one before it is foreshortened with the card. Either way the layer keeps its place in depth, so it still crosses other cards correctly.

More Transform 3D nodes after such an effect still move the card, and the effect keeps working on the camera's view. That is why parenting changes nothing about a layer's effects: the Parent Transform goes in just before Output, after all of them.

On things with depth (3D dots, shapes, a model), an effect always works on what the camera sees of them, wherever the Transform 3D sits. A Glow on a ball of dots keeps its size in pixels as the ball moves away.

Cards in depth ​

Cards are drawn with true depth, so cards that cross cut through each other, soft edges and see-through areas included: wherever two cards overlap, the nearer one is in front at every pixel.

Cards at the same depth stack in layer order, exactly as their 2D layers do.

A layer with an effect after its Transform 3D is drawn at its card's depth. Where its soft edges or a glow cross another see-through card, the overlap is close but not exact. Every 3D layer starts at z = 0, so a comp switched to 3D looks the same until you move something.

Up to eight see-through layers deep are exact at every pixel. Beyond that, the deeper ones are ordered card by card.

Depth inside one layer ​

A Merge crosses its 3D inputs by depth too, the same way layers cross: wherever two of them overlap, the nearer one is in front at every pixel, whatever their order in the Merge's list. So a ball of points turning around a flat disc that hides its far side can live in one layer:

Ellipse → Draw Shape → Transform 3D ───────────────┐
Point Grid (Sphere) → Transform 3D → Draw Points ──┴→ Merge → Output

A 3D input is anything drawn from points, shapes, text or a model with depth, or a picture that went through a Transform 3D on its way in. The rules are the layer stack's:

  • Inputs at the same depth stack in list order.
  • A 2D input keeps its place in the list, and splits the 3D inputs above it from the ones below it.
  • An effect between the Draw node (or the Transform 3D) and the Merge works on the camera's view and keeps its place in depth.
  • Only a Merge set to Normal crosses by depth. The other blend modes stack in list order.

A Transform 3D after the Merge moves everything in it together, and the inputs go on crossing by depth:

Ellipse → Draw Shape → Transform 3D ───────────────┐
Point Grid (Sphere) → Transform 3D → Draw Points ──┴→ Merge → Transform 3D → Output

That last Transform 3D places the disc and the ball together. With one there, the Merge's 2D inputs become cards at z = 0 and move with the rest, so a 2D input can start hiding what sits behind z = 0 even before you move anything. Switching a 2D layer to 3D in the layer stack does the same.

What reaches Output is one picture that still knows its depth, so the layer crosses other 3D layers correctly when they pass in front of or behind everything in it. A layer that passes between the merged things crosses that picture, and soft edges can pick up the wrong color there: for that, keep the things on separate layers.

The camera ​

Add Camera in the top bar adds a Camera layer. The topmost enabled camera layer is the comp's camera; without one, Caddis uses a default camera that shows z = 0 exactly as in 2D. Camera layers never draw.

Lens effects ​

Effects after the Camera node in a camera layer's graph are its lens: they work on everything below the camera layer in the stack, 3D and 2D, and leave layers above it alone, so a border or title above the camera stays clean. See Camera.

Views ​

In a comp with 3D layers, the Viewer gets a view menu:

ViewWhat it shows
CameraThrough the comp's camera — what renders
Custom ViewA free perspective view you orbit around the scene
Front / Top / LeftStraight-on views with no perspective

Working views never move the comp camera and never reach a render. In them, camera layers are drawn as their pyramid of view. Opening a comp with no 3D layers returns the Viewer to Camera.

The Viewer moves through a 3D comp the way 3D apps do:

GestureDoes
Alt + left-drag, or hold 3 + dragOrbit around what you're looking at
Alt + middle-drag, or hold 1 + dragPan
Alt + right-drag, or hold 2 + dragDolly in and out
Alt + wheelDolly in and out
FFrame the selected 3D layers (or all of them)
F1 / F2 / F3 / F4Camera / Top / Left / Front view

The three buttons at the Viewer's top right (Move, Dolly, Orbit) do the same when dragged, so you never need the keyboard.

Front / Top / Left are flat views: they pan and zoom, and don't orbit.

Navigating the Camera view leaves the camera where it is and switches to a Custom View that starts exactly where the camera stands. It orbits around the camera's Target.

Lock Camera to View ​

To move the camera itself, turn on Lock Camera to View (the padlock beside the view menu). The Viewer then looks through the camera, and every orbit, pan, dolly and F moves the camera layer: its Position and Target change, or its Rotation when Use Target is off. A rig of Transform 3Ds after the Camera node is taken into account, so the camera goes where you steer it. Each drag is one undo step. A value that already has keyframes gets a key at the playhead (with Auto-Key on), as with the other Viewer handles. A camera that follows a model file's camera can't be moved this way.

Moving and turning a card ​

Select a 3D layer (or one of its Transform 3D nodes). The handles work like the 2D transform gizmo:

  • Drag inside the card's outline to move it in its X/Y plane (in Top or Left view, the plane that faces you).
  • Drag the round handle above the card to turn it about Z, as in 2D. Shift snaps to 15°.
  • Drag along the red or green ring to turn it about X or Y. Shift snaps to 15°.
  • Drag a red, green or blue arrow to move it along just that axis.
  • Drag a corner of the outline to scale it, keeping its proportions; drag an edge to scale just its width or height (Shift scales both). Scale always follows the card's own sides.
  • Something with depth (a model, or points and shapes spread in depth) gets a 3D box instead of a flat outline. Drag a corner of the box to scale all three sides together, or the middle of a face to scale just its width, height or depth (Shift scales all three). Drag inside the box to move it.

A scale handle works by distance from the pivot: drag it twice as far from the pivot and the scale doubles, however the box is turned to you.

As in 2D, the size shows under the outline (width × height, plus depth for a box), and while you turn it the angle you've added shows on its pivot.

Local and World axes ​

The Local / World menu next to the view menu sets which axes the handles use:

  • Local (the default): the card's own axes. The arrows, rings and the drag inside the outline turn with the card, and with its parent when it has one, so you can slide a tilted card along itself.
  • World: the comp's X, Y and Z, whatever the card or its parent is doing.

Either way, a drag sets the node's own Position and Rotation. One drag can change several of their values at once: sliding a turned card along its own X changes Position X and Z, and a World turn can change all three Rotation values. Caddis keeps them as close to the old values as it can, so curves don't jump.

Parenting ​

Pick a parent in a 3D layer's or camera's Parent column in the timeline. As in 2D, this adds a Parent node and a Parent Transform just before Output, wired Position, Rotation and Scale, so you can put a Math or Vector Math node on a wire for an offset. The layer then follows everything its parent does in 3D. Parenting and unparenting never move the layer; unparenting leaves the Parent Transform holding where the layer is.

The Parent Transform sits just before Output on every kind of layer, and moves everything before it: a card, dots and shapes drawn in depth (they stay in depth), a model, or several things merged into one layer.

A parent can be any 3D layer or camera. A layer hands its children the placement of every Transform 3D along its chain: the ones on its points or shapes and the ones after its Draw node alike. A 2D layer has no place in 3D, so choosing one (a Null, say) switches its 3D on. Parent a camera to a Null to fly the whole rig at once.

The Parent Transform has no handles (its wires set its values); the layer's own Transform 3D does, and they move it along its parent's axes.

Motion blur and depth of field ​

A 3D layer with its MB switch on blurs along its 3D motion, and with the camera's motion, using the comp's shutter — cards, and points and shapes moved by a Transform 3D alike. Motion it takes from another layer counts too: its parent's, a published value its Transform 3D is wired to, a camera carried by a null. The camera's Depth of Field settings blur 3D layers by their distance from its Focus — through the comp camera only; working views stay sharp.

In After Effects ​

The Caddis effect in After Effects sees a graph's 3D through the comp's own camera, so it lines up with After Effects' 3D layers. Open Caddis Project turns cameras into After Effects cameras and cards into 3D layers, and Export Comp to Caddis brings them back. See Caddis in After Effects.