secondsun.lens

Stage 4 of The Second Sun — how the film is seen and heard. The film director calls Lens.tick(t, dt, frame) last of the four stages, because the lens has to see what the other three just made and the scene's ears ride the camera.

What it owns

• Thirteen camera set-ups driving FilmCamera's transform and Camera.fov.
• The post chain — a physical lens (defocus computed from the field of view the shot is cut for), a 180° shutter, a bounded iris, one colour table per act, bloom, a vignette and film grain.
• The sound design, in secondsun.sound.score, ticked from here.

Angles

A station is written as a compass bearing and an elevation, because that is how the rest of the film is written — the moon is at azimuth 305, the impact at azimuth 186. Bearing 0 is north (+Z), 90 east (+X), 180 south (−Z); elevation is degrees above the horizon; roll is clockwise in frame.

Lens.orient(bearing, elevation, roll) returns the quaternion the entity takes. It is a quaternion rather than an Euler triple because shot 9 points at the zenith and rolls, which is exactly where a yaw/pitch/roll triple gimbal-locks. Verified against camera.lookAt on all six axes and at the zenith.

Lens.aimAt(eye, target) gives the bearing and elevation from one point to another, so a station can name a thing in the world instead of an angle — which is what most of them do.

The stations

Shot 1 and shot 13 are one function called twice. Shot 13 replays shot 1's move exactly — same start point, same 1.4 m/s forward, same bearing, same lens, same focus rack. Every authored term is identical. eye.y is Film.seaLevel(t) + 0.22 in both, the same expression: at t=0 that is 22 cm above a sea at 0, and at t=88 it is 22 cm above a sea 11 m higher. Lens.stationIdentity(tau) prints both, term by term, with an equal block.

Shot 3 and shot 11 are likewise one function, so the footprint is framed identically before and after.

Marks

Lens.GEO holds where the lens believes the sand bar, the footprint, the channel, the forest edge and the wave front are. Every half second of film time the lens also looks for entities named SandBar / Herd / HatchlingFoot / Channel / ForestEdge / SurgeFront, and aims at those instead when they exist. So a station tracks what the shore and life stages actually stood, and falls back to the authored number until they do.

The lens as a lens

A full-frame sensor is 24 mm tall, so the focal length a vertical field of view describes is 12 / tan(fov/2). The station names a field of view and the defocus is computed from it, so the two can never disagree: the 26° lens of shot 12 is 52 mm and the 72° lens of shot 9 is 16.5 mm, and each defocuses the way that focal length and that aperture do.

The iris

Adaptation alone lags a 0.4 s white-out, so the film rides the iris as well as metering it. AutoExposure runs with minEV −7.5 / maxEV 4.0, speedDown 6.0 and speedUp 0.45 — it stops down fast and opens slowly, the way one does — and on top of that compensationAt(t) takes −3.1 stops into FLASH, gives 1.1 back as the scene settles, takes another −1.9 across the EJECTA ramp and returns 1.6 under the lid. Between the two, nothing in the film clips.

The grade

One colour table per act, baked with lut.fromGrade into /zero/source/secondsun/lens/ and bound to a single lut pass whose texture is swapped when the act changes. Act I is the only cool one and the only one with the saturation that far down; nothing after it goes back. The swap at FLASH is the one act change that does not fall on a cut, so amount dips to nothing across the 0.35 s the horizon is white and comes back on the far side.

Reading it back

Lens.report() — the live station, the focal length, the camera's actual pose, the iris, the defocus, the stock, the grade, the post chain and the whole mix. Lens.teardown() drops the chain and the sound sources so the next tick rebuilds them.