← Journal

Where AI fits in the work

Generative tools have changed what a small studio can produce, but not what makes the output good. A practical account of where they earn their place — and where they quietly cost more than they save.

AI-assisted campaign imagery

The interesting question about generative tools is no longer whether the output is usable. It often is. The question is which parts of a project they actually shorten, because the answer is narrower than the marketing suggests — and knowing the boundary is what keeps the work from looking generic.

Where they earn their place

Exploration. The fastest route from a vague brief to a shared reference point. Twenty directions in an afternoon is genuinely new, and most of them being wrong is fine — their job is to make the right one identifiable.

Volume around a fixed idea. Once the direction is locked, producing the variations a campaign needs — different crops, formats, seasonal versions — is where the time actually goes on most projects. This is the least glamorous use and the most valuable one.

Things that would otherwise not get made. Concept frames for a pitch, a mood film for an internal presentation, a product shot for a version of the product that doesn’t physically exist yet. Work that previously died at the budget conversation.

Where they cost more than they save

Anything requiring exactness. Type, logos, packaging copy, specific product details. The correction loop is longer than doing it properly the first time, and the result is usually a composite anyway.

Consistency across a set, unassisted. Left alone, generative output drifts — the light changes, the subject changes, the world changes. Holding a set together takes the same discipline as any other campaign: anchors, references, and rules agreed before generating anything.

Taste. The tools have no view on whether an image is right for this brand. They will produce something competent and characterless forever if nobody stops them. The bottleneck moved; it didn’t disappear.

The part that hasn’t changed

A tool that makes a thousand images cheap makes judgement expensive.

Every decision that mattered before still matters: what the brand is arguing for, what’s excluded, how restrained the palette is, whether the whole set holds together. Generative tools change the cost of producing a frame. They don’t change what makes the frame belong.

The practical version of this is unglamorous. Direction is written first, the same way it always was. References are locked. The generated material is treated like footage — raw, plentiful, and mostly discarded. What survives is chosen against the same criteria as anything shot conventionally.

That’s the whole method. The tools are faster; the standard is the same.

[More reading]

All posts