Two keys, a clock,
and substitutions nothing tells you about.

The straight fraction of elapsed time is only the starting point. It is passed through the departing key's Ease first, and there are five to choose from.

The panel calls it ease out of this key, which is exactly right and easy to read past. A key's ease governs the segment that starts there. It says nothing about how the camera arrives at it — that was decided by the key before.

So to soften the end of a move, you change the ease on the key you are moving from, and set it to out. Changing the destination key does nothing, which is the more natural thing to try and the reason this comes first.

Both were measured on a ten-second move from x=0 to x=100: at 9.9 seconds the camera was still exactly at zero, and at 10 seconds it was at 100. A held shot, then a single-frame snap.

They are interchangeable in effect and not in authoring. Cut is the one to reach for, because it describes the edit the way an editor would — "cut to this shot" — and it travels with the shot it introduces. hold is useful when you want the previous shot to sit still for a beat and the reason belongs to that shot rather than the next one.

With three keys at (0,0), (50,40) and (100,0), the straight-line halfway point of the first segment would be (25, 20). Measured, the camera was at (21.9, 22.5) — noticeably off the line, bowed by the pull of the key that comes after.

This is what makes camera moves look filmed rather than driven, and it is also why a camera can swing wide of something you framed it against. If a move clips a wall it never should have reached, the cause is usually the key beyond the one you were watching.

A Marker is not a small or invisible prop — it produces no geometry at all, and still produces none if you give it an asset path. It exists to be aimed at: you put one where an actor will stand, then point a camera key's look-at at it, and the shot is composed against a thing the game will never render.

The unrecognised case matters more than it looks. A scene saved by a newer version of the tool, carrying a prop kind this one has never heard of, opens as a grey box in the right place rather than dropping the prop. Your composition survives a version gap with a placeholder instead of a hole.

Unless it cannot. The base-game pack is opt-in and is off until you mount it, so on a fresh checkout every asset path falls through. What you get is the prop's Kind, and this is the part worth internalising: it is not a generic placeholder. A prop set to Sphere with an asset path falls back to a sphere, verified.

Which turns Kind from dead metadata into a decision. Once you set an asset path it is easy to stop thinking about the shape underneath — but that shape is what you and everyone without the game will actually see.

The environment one is the easiest to misread. Ask for a preset the tool does not ship — a typo, or one removed since the scene was made — and you get Clear Noon, which is also the default a brand-new cutscene starts on. A scene whose careful night lighting has quietly become midday is indistinguishable from a scene nobody set an environment on.

The material one is subtler still. A prop stores its material as the set's hash, not its name, because names in a real library are not unique — several sets share one. Storing a name would mean "the first set called this", so re-indexing the library could hand a saved scene a different material than the one it was composed with. The hash avoids that, and the cost is that a missing set gives you a flat colour with nothing readable to search for.

A parcel is a whole prefab place — the shipped duplex is hundreds of placements — staged by name under its own root. It is not in the prop list, and it should not be: a few hundred rows of building parts would bury the four props you positioned by hand. Swap the parcel and your props stay exactly where they were, verified.

A VFX prop carries a time and a duration, and no other prop does. That is the whole reason effects are a separate kind rather than another asset path: a muzzle flash happens on a beat; a crate simply is. A duration of zero means it never stops emitting — right for smoke drifting through a shot, wrong for an impact.

The runtime export serialises the cutscene whole — it is the model, written out. The bundle is a deliberately reduced projection: name, duration, and for each track its name, type and keys. It is a timing document, and a good one, but the stage is not in it.

Neither button says so. They log the same success and hand back a path, and the smaller file is only obviously smaller if you go looking. A quarter of the size is the giveaway.

That right-hand column is not hypothetical. A cutscene was built carrying every one of those five problems at once and validation returned clean — no issues, no warnings about any of them.

Which follows directly from tier two. The stage substitutes silently for everything it cannot find, so by the time validation looks, there is nothing to find wrong: the asset already fell back to a primitive, the preset already became Clear Noon, the ease name already became inout. A validator downstream of a silent fallback can only ever see success.

Note what drives the count. It is the cutscene's Duration, not the time of its last key — so the still frame you deliberately left at the end, from tier one, costs real frames and real time. If a render is taking longer than the action lasts, that is why.

The render is deterministic by design: the playhead is stepped to frame / fps and posed there, rather than played in real time, so the result does not depend on how fast your machine drew the previous frame. The same cutscene rendered twice gives the same frames.