Nova Studio

Spatial media

3D scene composition

Composing a place rather than a picture: what is in it, how it is lit, and what stays true when the camera moves to somewhere nobody composed for.

Composing a place

A picture is composed once, for one viewpoint, by somebody who chose that viewpoint. Every decision in it can be made in service of that single frame — and that is exactly why a beautiful still often falls apart the moment you move six inches to the left.

A place has to survive being looked at from angles nobody composed for. That means the things in it need positions relative to each other rather than positions in a frame, and the light has to belong to the room instead of to the shot.

The graph is where those relationships live. A scene node holds the place. View nodes draw from it. Changing the place changes every view of it, which is the only practical way to keep a room being the same room across a dozen shots built over a week.

On the director stage

Shots against one scene

Every shot in a sequence draws on the same scene reference, so the room is the same room in each of them without anybody checking it by eye.

Light as a shared input

Time of day and key direction belong to the scene rather than being restated per shot. This is what stops the sun moving between two cuts of the same afternoon — the single most common continuity failure in generated sequences.

Blocking as structure

Where the subject is in the space is a property of the shot, connected to the scene, rather than described from scratch every time and drifting a little with each description.

Coverage that agrees

A wide and its matching close-up share a parent, so they are coverage of one moment rather than two images of a similar moment.

A method that works

  1. Establish the place at its widest

    Generate the establishing view first. It carries the most information about layout and light, and everything else inherits from it.

  2. Fix the light before the angles

    Time of day and key direction, decided once. Changing the light after you have twelve views means regenerating twelve views.

  3. Derive views, do not describe them

    Each new angle connects to the scene and adds only what is specific to it. If you find yourself re-describing the room, the connection is missing.

  4. Check the pairs

    Look at any two views side by side and ask whether they are the same place. This is the only reliable test, and it takes ten seconds.

Generated, not geometric

For work where geometry is the point, build it properly and use this for the look. The two are complementary and the handoff is the intended path, not a workaround for a missing feature.

Questions

How many views can a scene realistically hold?

Enough for a scene; not enough for a walkthrough. Coherence degrades as you add angles, and the failure is gradual rather than sudden, so check pairs as you go rather than at the end.

Can I use a real photograph as the scene?

Yes, and it is often the strongest approach. A photograph as the scene reference gives you a place with real light that the derived views inherit.

What breaks first as a scene grows?

Usually small props and background detail, well before the architecture. Compose so that the things that matter are the things that are large and lit.

Related