A campaign, start to finish, with the dead ends left in

A campaign looks linear in retrospect and is not while you are in it. What follows is the actual sequence — brief, extraction, direction, production, failure, recovery, delivery — with the parts that wasted time included, because those are the parts worth learning from.

The specifics belong to one job. The shape repeats: roughly a fifth of the effort in understanding, a fifth in establishing the look, a fifth in production, and the remaining two fifths in the checking, fixing and delivering that nobody plans for.

Budget accordingly and the job is calm. Budget for production only and it is not.

Read the brief for hard constraints, stated negatives, and references with a note on what is being referenced. Turn it into one page and send it back for confirmation.

In parallel, build the brand file: palette, tone, permitted locations, guardrails, and a preservation list per product. Ask for reference photography — multiple angles per item — while doing it, because that request has a lead time and blocks everything downstream.

The confirmation reply is worth waiting for. Starting production against an unconfirmed reading is the most expensive risk in the job, and it costs one day to remove.

Produce one representative frame to a finished standard. Not the hero — something typical, whose lighting and treatment you want everywhere.

This is where the first real dead end usually appears. A treatment that read well on a board turns out not to survive the product: too dark to see the material, too stylised to keep the object honest, too busy for the lower third the layout needs.

Recognising that early is the whole point of anchoring. Discovering it after twenty frames means twenty frames to redo, and the temptation at that stage is to keep them and hope, which never works.

Lock the anchor, record its prompt and settings, and treat it as the specification for everything after.

Every remaining frame is sequence block plus scene block plus product block, with the anchor attached as a style reference. Assembled, not typed.

Run in batches and review in batches, against the preservation list and the guardrails rather than on impression. The check should refuse — wrong ratio, missing count, an attribute changed — before anything reaches a review folder.

Expect a hit rate well below what a single good frame suggests. Producing forty to select twenty is normal, and planning for it removes the panic when the first batch comes back mixed.

Bring every selected frame into one session and match them: exposure, balance, contrast, as a group. Small differences that were invisible frame by frame are obvious in a grid, and the grid is how the client will see them.

This is also the moment to catch the frame that does not belong. There is usually one — technically fine, subtly from a different world. It reads as an error in a set even though it would be fine alone. Replace it or cut it.

Half a day here does more for the campaign's coherence than any amount of additional generation.

Deliver with the folder structure, the naming convention, and the notes page. Include the masters and the recipes, even though nobody asked.

Then expect the round. Sort the notes into adjustments and structural changes, do all the adjustments first, and quote the structural ones separately. Most of what arrives will be adjustments, and delivering those quickly resolves a surprising number of the structural notes without touching them.

What makes this round cheap is entirely decisions made in week one: layered production, saved recipes, a written specification to check against. None of it is visible in the finished work, and all of it is why the job ended on time.