People keep asking whether AI can make images.
Of course it can make images.
That is not the interesting problem.
The interesting problem is that VFX is not image generation.

VFX is memory.

It is remembering what the client said two weeks ago.
It is remembering why version 014 was rejected.
It is remembering that the actor’s edge looked fine in a still frame but fell apart in motion.
It is remembering that the smoke element worked in one shot but cannot be reused in the next because the wind direction is wrong.
It is remembering that the director wants the creature to feel heavier, but the supervisor hates how overexposed it gets when the flashlight hits.
None of this fits neatly inside a prompt.

It lives in dailies.
It lives in shot notes.
It lives in Slack threads.
It lives in the compositor’s head.
It lives in the lighter’s head.
It lives in the coordinator’s spreadsheet.
It lives in the producer’s panic.
It lives in the shared memory of a team slowly learning what the show actually wants.

That is knowledge work.

Current AI is weak at that part.
Not because it is stupid.
Because it has no continuity of responsibility.

A compositor does not just make a pretty frame.
A compositor carries the shot.

They know what has already been tried.
They know which cheat will survive motion.
They know when the comp is wrong because of the grade.
They know when the CG is wrong because of scale.
They know when the plate is lying.
They know when a note should be followed.
They also know when a note should be protected from itself.

This does not mean AI is useless.
It means output is not ownership.

A tool can produce pixels.
But the shot still has to be carried by someone who remembers why those pixels matter.

That is not button pushing.
That is judgment under context.