ae-camera-3d

Camera and 3D

Any layer becomes 3D with ae_set_layer_props {"props": {"threeDLayer": true}}. It then has Z Position, X/Y/Z Rotation and Orientation, and responds to cameras and lights. Position becomes [x, y, z]: +Z is away from the camera.

The renderer

The comp's 3D renderer decides what 3D can do. Set it with ae_set_comp_settings {"settings": {"renderer": "…"}}:

The lens, in numbers

After Effects measures a camera's Zoom in pixels, not millimetres. The conversion, for a 36 mm-wide sensor:

zoom_px = comp_width × focal_mm / 36
focal_mm = 36 × zoom_px / comp_width
fov = 2 × atan(comp_width / (2 × zoom_px))

So on a 1920-wide comp: a 24 mm wide lens is zoom 1280, a 35 mm is 1867, a 50 mm is 2667, an 85 mm portrait lens is 4533. On a 1080-wide vertical comp a 50 mm is zoom 1500 — which is exactly what After Effects writes when you create a camera with the 50 mm preset, so the formula is the same one it uses.

Two consequences worth knowing before you place anything:

• A long lens kills parallax. If the move needs depth to read, use a

wider lens and a shorter dolly rather than a long lens and a big one.

• Moving a layer back by Δz and keeping its size means scaling it by

(zoom + Δz) / zoom. That is the whole trick behind flat artwork laid out in depth.

Build depth first

Space the layers in Z before any camera exists. Foreground at Z −400 to −200, subject at 0, background at +800 to +2000. Scale the far layers up so they still fill the frame, by the factor above. Parallax comes from these Z gaps, not from animating each layer separately.

A layout that reads, for a 1920×1080 comp with a 35 mm lens (zoom 1867):

A camera rig

Never animate the camera directly. Parent it to nulls so each move is one property:

ae_create_layer {"kind": "camera", "options": {"name": "Camera"}}
ae_create_layer {"kind": "null", "options": {"name": "Cam Dolly", "threeD": true}}
ae_create_layer {"kind": "null", "options": {"name": "Cam Orbit", "threeD": true}}
ae_set_layer_props {"layer": "Camera", "props": {"parent": "Cam Dolly"}}
ae_set_layer_props {"layer": "Cam Dolly", "props": {"parent": "Cam Orbit"}}

Put Cam Orbit at the subject's position.

Parenting order matters. After Effects keeps a child's look in the world when it's parented. A layer that is already 3D, parented to a rotated 3D parent, gets a counter-rotation and leaves the parent's plane: layers "slice" into each other. Parent while the child is 2D, then set threeDLayer: true. (ae_set_layer_props applies parent before everything else in the same call.)

Then:

Ease camera moves long and soft (landing ease 80–95, 20–60 frames), because cameras have weight. Add a subtle handheld feel on the Camera itself: wiggle(0.6, 6) on Position (see ae-expressions).

Numbers that read as deliberate rather than accidental:

Orientation aims, Rotation animates. Set a camera or layer's Orientation once to point it where it should look, and keyframe X/Y/Z Rotation for the move. Animating both, or animating Orientation across more than 90°, is how you get a flip halfway through that nothing in the values explains.

Turn motion blur on for camera moves. A camera move is the fastest thing in most comps, and without blur a push-in strobes on every edge.

A camera's transform paths are ["Transform","Point of Interest"], ["Transform","Position"] and so on. Its lens settings are under ["Camera Options", …].

Depth of field

A rack focus: key Focus Distance from the foreground layer's distance to the background's over 15–30 frames, eased. To keep focus locked on a layer, use an expression on Focus Distance:

length(thisComp.layer("Subject").toWorld([0,0,0]), toWorld([0,0,0]))

What each renderer costs you

The renderer is not only about features; it changes which of your other tools still work. Test with one ae_render_frame after switching, because the viewport and the render can disagree.

Whatever the renderer, effects are 2D: they run after the layer is placed in space, so a glow on a tilted 3D layer is a flat glow on the flat result. If an effect must live in the scene, put it on a precomp and make that precomp the 3D layer.

Lights and shadows

ae_create_layer {"kind": "light", "options": {"name": "Key", "lightType": "spot"}} (parallel, spot, point or ambient).

Layers only cast and receive shadows when their Material Options allow it:

ae_set_property {"layer": "Card", "path": ["Material Options","Casts Shadows"], "value": 1}
ae_set_property {"layer": "Floor", "path": ["Material Options","Accepts Shadows"], "value": 1}

Other material settings: Accepts Lights, Ambient, Diffuse, Specular Intensity, Specular Shininess, Metal, Reflection Intensity, Transparency, Index of Refraction. A classic setup: key light 100%, a dim ambient at 20–30% so shadows aren't black, and one rim light behind the subject.

3D models

With the Advanced 3D renderer (AE 2023+), ae_import_file a .glb, .gltf or .obj and add it with ae_create_layer kind footage. Models are lit by the comp's lights, or by an environment light (image-based lighting).

Camera shake and impacts

Shake the camera, not the layers, so the whole scene moves together and the parallax stays right. Put the marker-driven decaying shake from ae-expressions on Cam Orbit Position, scaled down (amplitude 4–10) for 3D.

What goes wrong in 3D

What needs a person

• 3D Camera Tracker (tracking real footage to get a camera) isn't

scriptable. Ask the user to run Animation ▸ Track Camera on the footage. Once it has a solved camera, everything above works with it.

• Choosing an environment light's image (HDRI) is set in the Effect Controls

or Project panel. Ask, or import the image and let the user assign it.

Check it

Depth and camera moves only read in motion. ae_review_motion over the whole move, and look for:

• layers that pass through each other or pop in Z
• edges of too-small background layers showing
• shadows that swim
• focus landing late

ae_render_frame at the start, middle and end confirms composition at each point of the move.

ae_proof helps less here than elsewhere, and says so: once a camera is in the comp it skips its space rules for 3D layers, because the sampler measures flat boxes, not what the camera sees (ae-motion-proof names the layers it skipped). Its timing, easing and integrity checks still apply to camera moves — a camera's keyframes are keyframes like any others.