AI Can Generate Images. It Doesn’t Know Why a Brand Should Exist
An AI-generated image can show a flawless coat and a garment that cannot be manufactured at the same time. Look at the fall of the sleeve, the position of a fastening or the way the fabric behaves. If these elements change during generation, the product in the image is no longer the one that will reach the store.
AI can visualise an idea quickly. It cannot tell the difference between an interesting variation and an error that changes construction, perceived quality or price.
We used to discuss ten images. Now we can discuss a hundred. Without precise criteria, we mostly end up with more images to reject.
Speed Helps When the Question Is Precise
At the initial research stage, AI can connect references, test combinations and make an idea visible before money and time are invested in prototypes or shoots. It also helps different functions discuss the same visual material. “Essential”, “protective” and “contemporary” can mean different things to design, merchandising and communication. An image exposes the disagreement.
Quantity does not guarantee difference. A study published in Science Advances found that AI assistance can improve an individual result while making the proposals as a whole more alike. The research concerned short-story writing, but the issue also matters to those who work with images: a series of variations can share the same idea of elegance, the same styling and the same references.
The problem often begins with the request. “A minimalist coat for a premium brand” describes a category, not an identity. It says nothing about who will wear it, how it should relate to the body, how long it should last, what will justify its price or which brand codes must remain recognisable.
A prompt can define lighting, palette and framing. The brand foundation must establish who the brand exists for, how it wants to be perceived, what it promises and which possibilities it needs to exclude. No stylistic instruction can replace those decisions.
An Archive Is Not a Folder of Images
A model can analyse and combine archive material. It does not know which elements built the brand and which belonged to a single season. It may repeat a frequently used colour without knowing whether it was an identity code, a temporary commercial choice or the result of a trend.
That hierarchy requires knowledge. A quiet detail may have held several collections together; a famous image may have been an exception. Treating them as equals produces a reference to the archive, though not necessarily an evolution of its language.
Heinz’s A.I. Ketchup campaign shows the issue from another angle. When DALL-E 2 was asked to draw ketchup, many of the generated images recalled the Heinz bottle and label. That association had already been built through decades of product, packaging, distribution and communication. The model reflected it.
For a less clearly defined brand, feeding the system a large number of images will not solve the lack of hierarchy. It can make the repetition of unrecognised or outdated codes more efficient.
When the Image Has to Match the Garment
The Mango Teen case is useful because it documents the process, not only the outcome. For the Sunset Dream collection campaign, the work began with real photographs of the garments. The model had to place them within generated images while preserving their characteristics.
Design, art direction, styling, data specialists and the photography studio worked together. The images were then selected, retouched and finished. The key constraint was the correspondence between representation and product: the campaign could not freely alter what the customer would buy.
This also clarifies the role of editing. The first check concerns the product. Does the fabric behave credibly? Is the seam compatible with the construction? Can the volume be achieved without changing weight and fit?
Next comes brand fit. A proposal can be technically correct and still sit too close to a competitor’s language. It may use archive codes too literally or alter them until they are no longer recognisable.
The commercial check follows. Can that shape be produced at the intended quality? Does the price remain coherent? Does the image promise something the real garment can deliver?
The order matters because a convincing image cannot compensate for a technically incorrect product. If the construction does not work, discussing the styling is premature.
Who Decides What to Reject
AI broadens the research and shortens the time needed to visualise a proposal. It does not take responsibility for discarding an attractive image when it does not belong to the brand or the product cannot support it.
The distinction between instruction and choice also appears in questions of authorship: in 2025, the U.S. Copyright Office found that prompting alone is not enough for an output to qualify for copyright protection.
If translating an image into a garment requires changing its proportion, material and price bracket, that image is of no use to the project. I would reject it.
August: From Copenhagen to AI
The month began in Copenhagen, asking when sustainability changes the product and becomes a brand language. Rebranding and brand creation shifted the focus to legibility and to the exclusions needed to define a point of view. The collection rail and trend analysis showed where positioning becomes concrete: assortment, proportion, material and the interpretation of shared signals.
Licensing, private label and Made in Italy tested how creative direction changes within different production and commercial models. The red carpet examined whether a point of view remains recognisable in public. AI closes August by asking what creative direction is still responsible for when images become easier to produce.
July had defined the theme “from brand DNA to market”. August clarified who must lead that transition: a Creative Director able to turn identity and vision into choices that can be verified in product, image and market.
Romina Tosi
Disclaimer
The views expressed above represent my personal interpretation of publicly available information and, like any interpretation, may be shared, debated, or challenged.
The information referenced comes from public sources available at the time of publication, including official documents, union communications, press articles, and materials accessible to anyone. I do not disclose confidential information or facts learned through privileged access. I do not reveal protected information, nor do I attribute unlawful conduct to individuals or companies.
Any reference to specific cases is intended solely to provide context and analyze dynamics affecting the broader industry. It is not intended to target individuals or particular businesses.
Observations regarding industrial strategies, financial decisions, and production models fall within the right to express opinions and commentary on matters of public interest. They remain personal assessments, not definitive judgments.
Not all companies operate in the same way. Alongside businesses that may deserve criticism, there are many others that work with seriousness, consistency, and long-term vision.
If you notice any errors or inaccuracies, please let me know. I will be happy to review and correct them where necessary.
The purpose of this reflection is to encourage discussion and debate, not to cause harm to individuals, companies, or organizations.
Comments