vawe-camera

Add a cameraMove block at the scene root, for example {"move": "diveIn", "start": 1.0, "dur": 2.2, "tx": 960, "ty": 420, "to": 1.6}. It bakes into data.camera at load via bakeCameraMove, and you can pass an array of specs to chain multiple moves.

Use a target reference like {"move": "diveIn", "target": "#cta", "margin": 0.1}. The baker resolves the layer's authored box, centers on it, and picks the largest scale keeping the box plus margin inside the frame. Layers need numeric x/y and a declared w/h or size.

Chained ease-in-out keyframes cause accelerate-stop-accelerate pulsing, known as the shaking zoom problem in rule #125. The generators emit interior ease:"linear" automatically; if hand-writing keyframes, set interior keys to linear and let only the final leg ease out.

No. A move holds its end pose forever until another move changes it; nothing resets at a cut or recipe seam. Author a return leg such as {"move": "slowPush", "to": 1} to come back to the normal frame.

Apply the plane modifier to separate layers in z, for example {"plane": -600} behind and {"plane": {"z": 240}} in front. Without depth differences every layer moves identically, so the move reads as a flat slide rather than a scene.

Targets need a measurable size: a text layer without w/h is refused rather than guessed, and [data-part] selectors require a live render so they cannot be resolved. Raw tx/ty values on the same spec always take precedence over a target.