Why a chat can't make a video

A chat is a line

Every message in a chat comes after the previous one, and the assistant's memory of your project is whatever fits in that line. That is fine for a conversation. A video is not a line; it is a small graph. The script depends on the sources. The voice depends on the script. The shots depend on the script and the voice's timing. The subtitles depend on the voice. Change the script and three things downstream are stale — and a chat has no way to know which three.

What "redo the whole thread" costs

In practice you paste the new script back in, ask for the shot list again, regenerate the voice elsewhere, re-time the visuals, and hope you did not lose the version of the hook that worked. The model did nothing wrong. The container lost the structure.

A canvas keeps the structure

Put each step on a board as a node and draw the dependencies as edges. Now the structure is visible and the tool can act on it: when the script node changes, the voice node knows it is downstream and can re-run on request; the shot nodes can stay put until you decide otherwise. You can look at the final video and follow the edges back to the sentence in the source that produced a given shot.

• Sources are read once and reused by every chat connected to them
• Each output is a node you can edit, regenerate or replace on its own
• Nothing downstream re-runs unless you ask
• Several people can work on the same board at once

Where the chat still belongs

Inside the canvas. Talking to a model is still the best way to say what you want; the canvas just gives that conversation a memory with edges in it. On Infica each chat is a node connected to the sources it may read and the outputs it produced, and there can be several on one board. The comparison with PoppyAI goes into how two canvas products differ in what they produce; the home page has the side-by-side with ChatGPT.