klik-cam

klik-cam — Camera Movement Craft

The camera is a character, not a viewport. Its energy is a branding decision made once per video, its moves are phrases with anticipation and settle like any other animation, and its repetition reads as lazy exactly the way a copy-pasted element does. Scope contract: klik-anim governs elements within the shot; this skill governs the shot itself. klik-anim's base laws bind every camera parameter — determinism (A0), thin-peak easing (A1), the pacing dial (C0.5), zero-static (A6, which the camera never satisfies on its own), and the render-and-look discipline (A9). Where this document says "A-law" or "C0.5", it means klik-anim's sections. v1 scope: the dial + the grammar laws. The deeper camera language — named pose-graph recipes, rig implementations, per-genre move vocabularies — is deliberately deferred to a later expansion.

C0. The Camera Motion Dial (1–5)

Set this once per video, during the brand/style phase — camera energy is a branding decision, not a per-shot whim. It replaces the moving-vs-locked binary with a continuous dial. Record the value in the project's style profile / spec (the method's SSOT §8 if in use), set in the same pass as klik-anim's pacing dial C0.5. The dial governs the camera only; the zero-static law (A6) still applies to elements at every setting — even at 1, something in the frame is always moving; it's the camera that holds.

Rules that scale with the dial:

• Anti-AI-camera (3–5): never a pure scale() / truck move — always pair zoom with a rotation axis or lateral vector; resolve every move into residual drift, never a dead stop.
• Monotone-spline poses (4–5): interpolate the pose graph so velocity is continuous but never overshoots a key — a direction reversal crests gradually, like a thrown ball. No bounce on the camera unless explicitly requested.
• Camera motion blur (3–5): fast segments need real shutter blur at the velocity peak (2–5f), landing sharp (@remotion/motion-blur CameraMotionBlur, or a shutter-angle frame blend). Missing it is the usual cause of judder.
• Depth pays off with movement: DOF + parallax do the most work at 4–5 (a moving camera separates the translateZ planes); at 1–2, depth is carried by static layering and cuts instead.
• Don't jump non-adjacent settings inside one video without a deliberate reason — a locked (1) editorial passage that suddenly hits a dial-5 orbit reads as two different videos. A one-step bump for a single hero moment is fine and effective.

CAM-1. Signature variation — no two scenes share a camera move

The camera column of klik-anim's device ledger (A14). A single rig applied uniformly makes a piece feel repetitive — "boring real fast" — even when every individual move is correct. A1 bans single-speed within a scene; this law bans single-signature across the piece.

• Direction, magnitude, and drift period must differ scene to scene. Two

scenes may share a rig; they may not share its numbers.

• Parameterising is not varying. A rig with variants whose every call

site passes the same explicit overrides ships identical moves — that exact dead-code failure ran for a whole session. Verify the variation reaches the render: compare frames from two scenes (A9), not the code.

• Drift on rests varies too — same period on every hold reads as a metronome.

CAM-2. Focus discipline — the focal plane sits on the words

In any DOF rig, focusZ follows the narrative target, and the narrative target is almost always the copy. A camera push that ramps focusZ off the type plane ships permanent blur on a hero line (a real 2.25px shipped defect). Camera moves change where the camera is, not what is in focus — unless the rack is the point, and then it lands ON the next target, crisp.

CAM-3. 3D-space hygiene

• **No grouping property on the space or its ancestors — filter OR

opacity < 1. Either one flattens the perspective** — every plane collapses to one (render-verified: a sample-stacking motion-blur wrapper's fractional opacity vanished the mid/near planes). Blur/warp/fade individual planes only (klik-anim A5 owns the blur budget).

• One perspective space per scene; planes mounted on translateZ, the

screen-flat reset plane for text/modals (A5).

• The camera is a transform on the space, elements are transforms on planes —

never mix the two on one node, or moves become un-reasonable-about.

CAM-4. Moves are phrases

A camera move obeys A1 like any element: anticipatory wind-up, thin-peak burst, decelerated settle — into residual drift, never a dead stop (3–5). Duration sits in A1's "camera phrase" band (1.6–3.6s), scaled by the pacing dial C0.5. Exits/entrances across a cut hand momentum to the next shot (match-motion; klik-anim A12's eye-trace law says where the eye must land).

Motivation leads; the camera reacts. The CAUSE of a camera move must be on screen ~10–14f before the move begins — a line emerging, a subject departing, a reveal starting. A camera that moves before its cause is visible reads as unmotivated and confuses the viewer ("why is the frame travelling?"). Anticipation is ELEMENT grammar (A2.2); reaction delay is camera grammar — the operator sees, then follows. Pair it with a front-loaded emergence on the motivating element (decisively out by the time the camera commits, soft arrival; the landing accent supplies the punctuation).

The named vocabulary — truck, pedestal, dolly, zoom, dolly-zoom, pan, tilt, roll, crane, whip, handheld, rack focus — with exact 2.5D implementations, the pan-vs-truck / zoom-vs-dolly distinctions, and the dial-gated move kit lives in `references/camera-moves.md` (each move render-verified in a labeled glossary demo).

CAM-6. The camera yields to the elements (the default behavior)

The camera's authority chain is elements > composition > music. The music sets the frame's TOTAL motion energy; the elements — already authored to the music — have first claim on it; the camera is sidechained to their motion and claims only the residual. This is the DEFAULT for every music-driven piece; the user may override it (more camera, less, or off) — an override is a recorded spec decision like any dial value, never a silent drift back.

• Sidechain, not switch. Element motion ducks camera drift with a fast

attack (~3f) and a slow, eased release (~24–36f) — the camera parks while elements phrase and swells back as they settle. Never frame-instant inverse coupling: that stutters the camera (CAM-4 violation).

• The C0 dial is the camera's claim multiplier on the residual — not a

mandate to move. Locked ≈ ×0, Anchored ≈ ×0.2, Motivated ≈ ×0.5, Flowing ≈ ×1, Perpetual ≈ ×1.5. The dial stays a branding decision; the sidechain decides moment-to-moment amount within it.

• Accents co-fire; drift yields. The duck governs sustained motion only.

On rank-1 beats (bar-aligned kicks, hot sections) the camera hits WITH the element impact — amplifier, not competitor — mostly on depth axes.

• Complementary axes. The camera never rides the axis the elements are

using: elements own the frame plane → the camera's residual lives in depth/roll. Same-axis stacking cancels or doubles element motion unpredictably.

• Gaze-anchored, margin-capped. Drift is defined relative to the focal

element's home and is mean-reverting; its amplitude is capped by A11's margin math so the focal centroid never leaves the inner pocket (~8% of frame). "Center" is defined by the subject — the camera cannot walk the boss off its composition. Which frame the cap is measured in follows the composition model (klik-anim references/layout.md): normally the video frame, but under the accretion artboard it is the reading pocket — the camera travels the canvas beneath a pocket that itself stays put, so the boss never leaves the pocket even though the canvas leaves the frame. A camera that walks the active run out of the pocket is the violation; one that walks spent canvas out of frame is the model working.

• Holds are the camera's turn. During designed holds (reading time) the

elements drop to micro idle and the camera provides the life — the slow push on a held line. One rank-1 mover at a time (klik-anim A10) — and the camera counts as a mover.

Implementations live in references/camera-music.md as presets: A · Hold & Glide (the default — camera holds with breathing only, then one composed zoom/slide per musical seam, landing on the downbeat; scenes laid out in world space so the travel IS the scene change) · B · Reactive Shake (the sidechained drift+hit rig — field-judged too twitchy as a standing behavior; opt-in, scoped, recorded) · C · Element-Follow (roadmap). Preset choice is the user's; A is what you reach for unasked.

CAM-5. The camera hears the music too

Same music map as klik-anim §A16, camera scale — the camera moves on the music's big body (bars, sections, drops), never its surface detail:

• Energy tier maps to activity within the dial.* Tier 5 is the dial's

ceiling (the one-step hero bump lives here, on the drop); tier 1–2 is its floor — drift only, even at dial 4.

• Signatures change at musical seams. CAM-1's scene-to-scene variation

lands at section boundaries, not at arbitrary cuts — a new section sounds different, so the camera moves differently.

• Bass impacts are camera hits: a small kick/push with shutter blur,

landing sharp on the transient. Swells are pushes/pulls whose ramp matches the swell's rise. Breakdowns settle into drift.

• Never mickey-mouse the hi-hats. Micro-percussion is element territory

(A16's blinks and ticks); a camera that twitches on every hat reads as nervous, not musical.

• Implementation recipes live in `references/camera-music.md` (this skill

folder) — load it when building a music-driven camera: the render-verified rig shape, per-section signature crossfades, the bar-kick filter, velocity shutter smear, and the sample-stacking-blur-flattens-3D trap.

CAM CHECKLIST — run before every render with any camera movement

1. Dial value set with the user (never defaulted), recorded in the spec, applied consistently — no unmotivated non-adjacent jumps?
2. At dial 1–2: camera truly holds (≤0.5% creep) — no accidental drift? At 3–5: every move pairs zoom with a rotation axis or lateral vector — no pure scale/truck?
3. Every move resolves into residual drift or a motivated rest — never a dead stop (3–5)?
4. Shutter blur present at every velocity peak (2–5f), landing frame sharp?
5. Camera signatures differ scene to scene (direction, magnitude, drift period) — verified in rendered frames, not parameters (CAM-1)?
6. focusZ on the narrative target (the copy) through every move — no hero line blurred by a push (CAM-2)?
7. No CSS filter on any ancestor of the 3D space; blur applied per plane only (CAM-3)?
8. Monotone-spline interpolation at 4–5 — no camera bounce, no velocity discontinuity at pose keys?
9. Camera follows the music map (CAM-5): signature changes at section boundaries, hits on bass impacts with blur, activity tracks the energy tier — and no camera twitch on micro-percussion?
10. Camera yields to the elements (CAM-6): drift ducked during element phrases (fast attack, slow release), focal centroid inside the inner margin the whole piece, no same-axis competition, accents co-fired only on rank-1 beats — or a recorded user override says otherwise?

Companion layer: `klik-anim` (element craft — its A-laws bind the camera too) · Remotion's official skills (mechanics) · the project's style bible (which dial value fits the brand).