Case Study: Campaign Key Visuals with AI — Before and After, With Numbers

Video gets the headlines, but most of an FMCG campaign is still flat. The billboard, the in-store poster, the shelf wobbler, the social grid, the packaging-adjacent banner nobody remembers approving — all of it hangs off one asset: the campaign key visual.

I’m the in-house AI creative director at a major FMCG company, and key visuals are where my AI pipeline quietly earns more money than the films do. Not because a single KV is expensive — because a campaign needs the same image redrawn eighteen different ways, and that multiplication is exactly what traditional production charges for and AI production doesn’t.

This is the breakdown of one real KV project: what the process looked like before, what it looks like now, what it cost, and where it nearly fell apart. As always with in-house work, the brand stays anonymous. The numbers don’t.

What is a campaign key visual, and why is it a perfect AI use case?

A key visual is the one master image a campaign is built around — hero product, world, mood, message, all in a single frame. Once it’s approved, it gets adapted: aspect ratios, retail formats, seasonal variants, regional language versions.

That structure is why KVs are arguably a better AI use case than video. Three reasons.

First, stills are a solved problem in a way video is not. The consistency issues I wrote about in brand consistency in AI video — drift across frames, packs reinventing themselves mid-shot — mostly disappear when there’s only one frame to hold.

Second, the exploration economics are absurd. A photoshoot gives you one art-directed setup per few hours of studio time. A generation pipeline gives you a hundred candidate worlds before lunch, and the craft moves to selection.

Third — and this is the money part — adaptation is nearly free. In the traditional model, every additional format is a line item. In an AI pipeline, the marginal cost of format nineteen is an afternoon, not an invoice.

What did the “before” look like?

Before this pipeline existed, a campaign KV at our scale followed a familiar script. Concept and moodboard, agency or studio quote, a photoshoot day (product stylist, photographer, lighting, set), retouching rounds, then a separate adaptation phase where a design team rebuilt the master into every required format.

Quantified from our own history and market rates: three to six weeks from brief to final adaptation set, and a budget anywhere from $5,000 to $30,000+ depending on how cinematic the world around the product needed to be. A change of creative direction after the shoot — different season, different mood — meant, effectively, a second shoot.

The deeper cost was rigidity. When the shoot is the expensive, unrepeatable event, every decision gets front-loaded and defended. Nobody explores after the shoot day; they salvage.

What did the “after” look like in numbers?

Here is the same class of deliverable through my current pipeline: one seasonal campaign, one master KV, full adaptation set.

Two numbers need context.

The ~90% discard rate is higher than my video rejection rate, and that’s by design. Stills are cheap to generate, so exploration widens. Most discarded frames weren’t broken; they were merely fine. “Merely fine” is exactly what a key visual can’t be, because this one image has to survive being blown up to billboard size and shrunk to a thumbnail. The discard pile is the quality system, same as it was in my food commercial case study — anyone selling you AI visuals without a discard budget is planning to ship the mediocre middle.

The 18 adaptations are the real business case. Hero 16:9, vertical, square, three retail formats, in-store POSM sizes, and regional variants — produced from two approved masters in under a day and a half, because the world of the image could be re-generated and re-framed instead of being cropped to death. In the before model, this phase alone had its own timeline and its own budget line.

How do you keep one image on-brand across 18 formats?

The honest answer: the pack never comes out of the generator.

The product itself — pack, label, logo — is handled through a reference-and-compositing pipeline, not left to the model’s imagination. The AI builds the world: light, surface, atmosphere, appetite cues. The pack is placed, corrected, and locked with old-fashioned retouching discipline. That’s the “2 compositing passes” row in the table, and it’s non-negotiable in FMCG, where a label that’s 5% off-model fails legal review, not just taste review.

The rest is boring rigor: one style bible per campaign (palette, grain, lens language, lighting logic written down before generation starts), and every adaptation checked against it. In one of our projects, the vertical formats kept drifting warmer than the hero — pleasant in isolation, obvious the moment the formats hung side by side in a store. Catching that is not an AI skill. It’s an art direction skill pointed at AI output.

What did it cost compared to a photoshoot?

I’ll frame this the way I framed the film comparison, because the honest version isn’t a single triumphant percentage.

Against the $5,000–30,000+ photoshoot route, this project’s total cost — my time plus the tool stack — landed under one-fifth of the comparable quote, for a deliverables list (two masters plus eighteen adaptations) that the traditional model would have priced as two separate phases.

But the sharper difference is the shape of the risk, not the size of the bill. A photoshoot concentrates the entire budget into one unrepeatable day; if the approved concept turns out to be wrong on the shelf, the money is spent. The AI pipeline spends nothing unrepeatable. When the brand team asked for a mood shift after seeing the first shortlist — warmer, more domestic, less studio — it cost one regeneration cycle inside the same four days. In the before world, that note arrives after the shoot, and it costs a shoot.

What still needs a photographer?

A case study without a limits section is an ad, so here are the limits.

Real product photography still wins when the asset must be forensically literal — packshots for e-commerce listings, anything where regulation requires the photographed product, and textures where appetite lives in micro-detail the brand won’t compromise on. I still commission straight packshots; they feed the compositing pipeline as reference.

Human faces at billboard scale remain a judgment call. Close-up emotional casting is doable but demands a discard rate even my table doesn’t capture, and in some markets the uncanny-valley risk to the brand isn’t worth the savings. I design around faces more often than I fight them.

And the master-KV decision itself — which of 120 frames carries a campaign — is the least automatable hour of the whole process. That hour is what the invoice is actually for.

What would a KV set like this cost your brand?

Scope decides, not secrets. A campaign key visual set — master plus full adaptation package — is work I price inside the campaign tier at $3,000–4,500 when bundled with video, or below that as a standalone stills project. A single hero spot with a key visual sits at $1,500–2,500. The ranges are public on my services page, because hiding prices behind a discovery call is a habit I don’t keep.

If your volume justifies building this pipeline in-house instead of hiring me, I’ll tell you that in the first conversation — I built one in-house myself, so I know exactly where that threshold sits.

FAQ

How much does an AI campaign key visual cost in 2026?

From a professional studio: roughly $800–3,000 for a master KV with an adaptation set, versus $5,000–30,000+ for a photoshoot-based campaign master. In-house marginal cost, once a pipeline exists, drops well below that.

How long does an AI key visual project take?

Four working days is my standard for brief to full adaptation set: two days of direction and exploration, one day of selects and brand review, and the rest for compositing, corrections, and formats. Bring one decision-maker and usable brand assets, and the four days hold.

Can AI keep the pack accurate across all the formats?

Not on its own — and any yes without a caveat is a sales pitch. Pack accuracy comes from a reference-and-compositing pipeline layered on top of generation. The AI builds the world; the pack is placed and locked by hand. Budget for that pass; it’s what makes the set shippable.

Do you replace product photography entirely?

No. Literal packshots, regulated imagery, and some micro-texture work still go to a camera. The AI pipeline replaces the expensive world-building around the product — the set, the atmosphere, the eighteen adaptations — not the product record itself.

I’m Mohammad Nasravi — in-house AI creative director at a major FMCG company. This before/after is how key visuals actually get made on my desk, limits included. If your next campaign needs a master and every format that hangs off it, see my services and pricing or get in touch — I reply within 24 hours.