What AIGC Production Actually Is
Two capabilities separate a viral demo from a shipped campaign. Here is what each one looks like when real brands run it.
Everyone has seen the demo. You type a sentence, wait a few seconds, and a photorealistic image appears. It is a genuinely impressive party trick, and it is almost nothing like what a brand actually needs on a Tuesday when a campaign ships Friday and finance is watching the number.
The demo makes one image. Production is a different animal. A brand does not buy an image, it buys a hundred assets that look like one campaign, cleared for use, sized for every placement, and working in six markets without reading like a clumsy translation in five of them. Once you have watched a real program run, the whole conversation shifts. Whether AI can make a picture is settled. What happens between that picture and a shipped campaign, and who is accountable for it, is the part nobody has automated.
Strip away the noise and AIGC production comes down to two capabilities. One is making every asset different on purpose. The other is making thousands of assets stay the same on purpose. They sound like opposites. In practice they are the two halves of the same craft, and the best real-world examples make the split easy to see.
The two things production actually does
| Mass individualization | Versioning at scale | |
|---|---|---|
| The goal | Every asset uniquely different | Every asset consistently on-brand |
| What varies | The design, per person or unit | Format, language, market, placement |
| What holds | The brand system underneath | The core idea across every cut |
| Classic failure | Variety that goes off-brand | Consistency that dies under volume |
| Who owns the outcome | A person, not the model | A person, not the model |
Both columns run on the same generative tools. Only one thing decides whether the output is usable, and it is not the model. It is the direction on top of it.
Making everything different, without losing the brand
Go back to 2017, before the current wave, for the cleanest example anyone has produced. Ferrero wanted to make the point that every Nutella customer is one of a kind. So the packaging became one of a kind too.
Working with Ogilvy Italy, Ferrero used a custom algorithm drawing on dozens of patterns and color palettes to generate seven million unique jar designs. Every jar sold through Italian supermarkets carried a different label. The run sold out in one month.
Read that again. Seven million labels, no two alike, every one still unmistakably Nutella. That is the whole trick of mass individualization. The variation is enormous, but it happens inside a brand system tight enough that a shopper never once thinks "that isn't Nutella." The algorithm made the combinations. People built the box those combinations were allowed to live in, the patterns, the palette, the rule that the logo never moves. Without that box, seven million unique labels is not seven million pieces of art. It is seven million ways to stop looking like yourself.
The same instinct shows up in the campaigns brands run now. Coca-Cola's Create Real Magic handed people a library of its own icons, the contour bottle, the script, the archive polar bears, and let them generate original artwork on top. The best entries went to billboards in New York and London.
Coca-Cola built Create Real Magic with OpenAI, combining GPT-4 and DALL-E with a controlled library of official brand assets, so that whatever a user generated was constrained to elements the brand had already approved.
Notice the pattern. Coke did not open the floodgates and hope. It fenced the playground first. The generation was open, the brand assets were not, and that boundary is the entire reason the output was usable instead of a liability. Individualization at scale is not "let the machine do whatever." It is deciding, with some care, exactly which things are allowed to change and which never are.
Making thousands of things stay the same
This is the half most marketers feel in their day, and it is where the numbers get hard to argue with.
A modern campaign is not one ad. It is the same idea cut for a vertical story, a square feed post, a banner, a product tile, then localized for language, then adjusted for legal in each market, then resized again for the next platform. The count climbs fast.
One analysis of creative operations put current global campaigns at 25 to 80 or more asset variants per market launch, and noted that a single Coca-Cola global campaign in early 2025 ran across 27 markets, each needing adapted visuals, localized copy, adjusted disclaimers, and format-specific cuts.
Do that by hand and the wheels come off. This is exactly where AIGC production, run properly, stops being a novelty and starts earning its keep. A few real numbers from brands that built the workflow instead of buying a toy:
| Brand | The job | The result |
|---|---|---|
| Shopify | One concept across 25 sizes, 7 languages, 25 variations | 4,375 assets in 12 to 24 hours |
| Samsung | From a single brief to approved retail assets | 500 approved assets in one hour |
| MegaFood | Refresh a full set of product-page assets | 1,100 assets in under four weeks |
| Zalando | Localized model photography for global markets | Production time cut by around 60% |
Figures as reported for Shopify's creative versioning program, and by Rocketium on Samsung and MegaFood creative automation, and by Zeely on Zalando's localized photography.
The Shopify number is the one to sit with. 4,375 finished assets in under a day, from work that used to take weeks. That is not just a faster tool. It changes what a small team can even attempt, because when production stops being the bottleneck, a failed test costs you an afternoon instead of half a month, and people start trying bolder things.
Where it goes wrong, honestly
Numbers like those make it tempting to stop and let the whole thing sound frictionless. It isn't, and pretending otherwise helps no one.
Coca-Cola's fully AI-generated holiday film drew real criticism for feeling emotionally flat, even as it made the technical point. Toys R Us put out an AI-generated brand video that got a public backlash. In both cases the tools worked exactly as advertised. What was missing was the judgment about whether the output actually moved anyone, and whether this was the right job to hand the machine at all.
That is the recurring lesson under every one of these stories, the good and the bad. The model is capacity. It generates, versions, and localizes, all of it faster than any team could manage by hand. What it cannot do is decide. It cannot tell you the idea is worth making, that this frame is the one, that the tenth variation drifted off-brand and needs to die, that a real emotion is missing and no amount of resolution will fake it. Capacity without judgment gives you volume you cannot use. Judgment without capacity gives you three nice images and a missed deadline. The work is holding both.
How we run it at HubStudio
We built the studio around that exact split. Senior creatives lead every concept. AI carries the production volume underneath them. The generation is the engine, the direction is the product, and we are clear about which is which because getting it backward is how content programs quietly fail.
How you buy it is simple. Hand us a brief and take delivery. Run your own team on a platform we stand up for you. Or bring us in to direct while your people execute. Same studio, three ways in, matched to how much you want to own.
And the work itself lands in one of three modes, chosen per job. We shoot, when a real camera is the right call. We generate, when the piece lives entirely in AI. Or we shoot and then adapt, capturing a hero the traditional way and extending it into the fifty variations a modern campaign needs, across formats, languages, and cultural context, so the version that lands in one market does not fall apart in the next.
Across recent engagements, HubStudio has moved first-round approval rates from 22% to 78%, cut production costs by roughly 60%, and lifted usable asset output around sevenfold on comparable briefs.
None of that comes from the model on its own. It comes from putting real production capacity behind work a person has already judged to be right. That is the slower thing to build, and it is the reason the output holds up when it matters.
See it on your own brief
The fastest way to judge any of this is on your own work, not ours. Send us a live brief. We will walk you through real brand cases, and we will produce a small set of test visuals against your actual requirements, at no cost and no obligation, so you can see the direction and the output before anything is on the line.
