What this is

Most writing about generative video explains how to make one impressive shot. Almost none of it explains what happens on shot forty, when the character’s jacket has changed colour three times, nobody remembers which version the client approved, and the shoot is on Thursday.

That gap is what this is for.

What gets published here

Working methods, not announcements. Character continuity across a full scene. What previz built on generative models actually replaces, and what it doesn’t. How a two-person team runs pre-production without a producer chasing versions in a group chat. What a shot costs, in real numbers, on a real project.

Every article here carries at least one thing that came out of an actual production: a screenshot from the workspace, a number we measured, or a failure we documented while it was happening. If a piece could be written without access to those productions, it doesn’t get published.

Why that rule exists

The tooling around generative video moves faster than anyone can write about it. Models arrive, get adopted, and get deprecated inside a single quarter — Sora’s API shutdown in September 2026 removed a step from pipelines that teams had built over a year.

What survives that churn isn’t knowledge of a particular model. It’s the production discipline around it: how you structure a project so that swapping the model doesn’t mean rebuilding the work. That discipline can only be described by people doing it, and that’s the only thing worth writing down.