Where people stand, and which way they face

Blocking is a still-frame decision

Blocking means where people and objects are placed, and how they move through the space. On a set it is worked out with actors. In generative video there are no actors, so blocking is decided entirely in the first frame, and that means you can actually control it, which is unusual in this medium.

The things to decide before generating:

• Where the subject is in frame: left third, centre, right third.
• Which way they face: toward camera, away, left, right.
• What is in front of them and behind them.
• Where they will be at the end of the shot, if they move.

A model asked to animate will otherwise default to what stock footage does: subject centred, facing camera, standing still.

The 180-degree line

The one rule from editing that generative work breaks constantly.

Imagine a line drawn through two people in conversation. Keep the camera on one side of that line and A always appears on the left looking right, B always on the right looking left. Cut across the line and both appear to be looking the same way, and the audience abruptly cannot tell who is talking to whom. The technical name is crossing the line; the experience is a small moment of confusion the viewer cannot explain.

On a set, a script supervisor prevents this. In generative work, nothing prevents it, because each shot is generated independently with no knowledge of the others.

So you enforce it in the stills. Write the direction into every shot on the list:

s04  MCU  Priya, screen left, looking RIGHT
s05  MCU  Arjun, screen right, looking LEFT
s06  MCU  Priya, screen left, looking RIGHT

Then build each still to match, and check the stills side by side before generating any of them. Two minutes of checking saves a scene that cannot be cut.

Movement direction across a cut

The same rule applies to anything moving. If a car exits frame right in shot one, it should enter frame left in shot two, because it is continuing in the same direction through an imagined space. Enter frame right and it appears to have turned around.

This matters in generative video more than usual, because you are frequently cutting between shots that were never in the same place at all. Consistent screen direction is a large part of what convinces an audience that a set of unrelated generated clips is one continuous journey. It is, genuinely, the cheapest continuity tool you have: it costs nothing but a note on a shot list.

Staging in depth

Placing people at different distances from camera (one near, one far) rather than side by side does three things:

• It creates the depth planes that make camera moves work, from the composition lesson.
• It lets one shot carry two subjects without either being a tiny background figure.
• It gives the model a foreground element to anchor the frame, which reduces drift.

A figure in soft foreground at the edge of frame, even just a shoulder, stabilises a shot noticeably. This is also how over-the-shoulder shots work, and why they are so useful: the shoulder is a large simple shape that holds, and it hides half the identity problem.

Entrances and exits

Asking a subject to walk into or out of frame is asking the model to invent them mid-clip or dispose of them convincingly. Entrances are harder than exits: something has to appear from a region that does not exist.

Two workable approaches. Put them partly in frame at the start, so they are entering rather than appearing. Or generate the exit only and let the edit handle the entrance by cutting to a shot where they are already there. The cut is free; the invention is not.

Eyelines as a directing instrument

Where a subject looks tells the audience what the next shot contains. A character looking off-frame left, then a cut to a window, and the audience has understood a spatial relationship that was never depicted.

This is enormously powerful in generative work because it lets two completely unrelated clips become a cause and an effect. Generate a person looking down-left. Generate a dog. Cut them together. The audience constructs a room that neither clip contains.

Specify eyeline explicitly in the instruction (“looking down and to the left, out of frame”) because a model left to itself will point the eyes at the lens, which is the one direction that breaks this effect.

Today: take any two-person scene on your list and write screen direction on every line. Then lay the stills side by side and check that nobody has swapped sides.