An AI ad in a day: brief in the morning, cut overnight

A short advertising piece can genuinely go from brief to cut in a day. It happens when the brief is clear, the product references exist, the approval is one person, and the shot list avoids what the tools cannot do.

Remove any of those and the day becomes three. The single most common cause is not production speed — it is waiting for a decision, or discovering at hour six that a required shot was never going to generate.

What follows is an hour-by-hour shape. The hours are indicative; the ordering is not, because each stage removes a risk that would otherwise surface later.

Read the brief for hard constraints, stated negatives, and any reference with a note on what specifically is being referenced. One page, sent back for confirmation.

Collect the product references at the same time — real photographs, multiple angles, unretouched. Without these, product shots cannot be done properly and no amount of prompting substitutes.

Then write the shot list, and check every row against what the tools reliably do. Mark anything involving hand interaction, rotation, readable type or several interacting figures. Decide now how each of those will be handled: composited, reframed, or cut from the list.

Choose the shot that carries most of the look — typically a medium shot in the main setting — and work on it until it is genuinely right. Twenty attempts here is a good investment; everything downstream inherits from it.

This is also where a treatment that read well on paper reveals whether it survives the product: too dark to see the material, too stylised to keep the object honest, no room in the frame for the end card.

Lock it. Record the prompt, the seed, the references and the settings. This frame is now the specification and the style reference for every remaining shot.

Every remaining shot is sequence block plus scene block plus shot row, with the anchor attached as a reference. Assembled from the list rather than typed individually — this is what keeps continuity.

Generate in batches and review in batches, against the preservation list and the brand guardrails before assessing quality. A check that refuses on a wrong ratio, a missing count or a changed product attribute is worth more than a careful eye at hour seven.

Expect to generate roughly twice what you keep. Planning for that removes the panic when the first batch comes back mixed, which it will.

Cut to the shot list. Because the list was structured and the shots were generated against one anchor, the assembly is mostly placing clips rather than searching for what works.

Then grade every shot together in one pass. Individually fine clips will not match, and matching them as a group is half an hour that does more for coherence than another twenty generations.

Type goes on last, set properly in a design tool over the space you left for it. Never generated — and the space has to have been planned at hour two, which is why it is on the shot list.

Approval by committee. One decision-maker available on the day, or it is not a one-day job. Say this at the briefing rather than discovering it.

Missing product references. Blocks everything downstream and has a lead time you do not control. Request at hour one.

An impossible shot kept on the list. Consumes hours confirming what a two-minute check would have told you.

A treatment chosen without testing it on the product. The anchor frame exists to catch this early; skipping it moves the discovery to hour eight.

Actual generation speed is rarely the constraint. The schedule fails on decisions, inputs and plans — which is why the first two hours carry so much of the weight.