AI design: the work between an image and a brand

You receive a presentation showing a logo on an elegant box, an illuminated shopfront and a phone screen. All three images seem to belong to the same brand. You approve the proposal. Later, you need to print a small label, add a product to the shop and prepare a campaign. That is when you discover that nobody decided how the identity works beyond the presentation images.

The name becomes unreadable on the label. The typeface lacks a suitable weight for prices. The purchase button disappears into the background. To produce another application, someone has to ask the tool for another image and hope it resembles the previous ones.

The gap between a convincing image and a usable brand reveals the competence of the person selling the design. Generating a proposal does not demonstrate an ability to direct it, assess it or turn it into a system. When someone who cannot do that work sells it as a professional identity, the client is buying a responsibility the supplier cannot fulfil.

Josué develops this criticism in engineering when he discusses the confusion between generated output and professional competence. In design, appearance can conceal the same failure. The difference becomes visible when we need to explain a choice, resolve a conflict or produce the next piece of work.

The first decision comes before choosing an image

If the initial request is simply “make a premium brand for a ceramics business”, almost everything remains undecided. What does it sell? To whom? Why should someone choose it? Where will it appear? Who will need to work with the files after delivery?

A tool can fill that gap with a plausible composition. But a word such as “premium” does not define the relationship between price, material, audience and buying experience. It might produce black and gold when the business needs to show the quality of its glaze, or delicate lettering that disappears in the stamp used in the workshop.

My first requirement comes before aesthetic selection: I want to understand the problem each decision is meant to solve. If the explanation for the colour, typeface and photography is always “it looks more sophisticated”, direction is missing. There is a visual preference, but there is no design argument we can assess yet.

This work also protects originality. Repeatedly requesting something “more modern”, “cleaner” or “more exclusive” does not clarify what makes the brand particular. Its difference needs to come from something the business can support: a production method, a relationship with a place, a service or a material choice.

An example: from a ceramics workshop to a visual direction

Consider a hypothetical case, created to work through the reasoning. A small workshop makes tableware in limited runs. It sells individual pieces to consumers and sets to restaurants. Variations in the glaze are part of the product, but dimensions and lead times need to be clear. The team wants to update its packaging and introduce a trade ordering area.

I would express the intention in plain terms: reveal the character of the material without making the workshop seem unable to fulfil an order. That creates a useful tension. We need sensitivity in the way we present the pieces and precision in the way we present the offer.

Imagine two directions. The first uses an irregular symbol, handwritten lettering and photographs crowded with objects. It might convey intimacy and handmaking, but makes collections harder to distinguish. The second uses a strict grid, white surfaces and cut-out products with no context. Comparison becomes easier, but the workshop may lose what makes it recognisable.

I would develop a third hypothesis: an orderly composition, photography close enough to reveal the material, and a graphic element drawn from an actual characteristic of the pieces. I would not choose a cracked texture merely because it looks handmade if the workshop is distinguished by consistent production and smooth finishes.

This is a proposed direction, still to be tested. The visual hierarchy, meaning the order in which the eye encounters information, must distinguish the collection name, dimensions, finish and price. An identity can have personality without asking buyers to decipher the catalogue.

The direction has to survive the label and the catalogue

This is where work omitted by a presentation image begins. The symbol needs versions suited to the agreed applications. A composition that works on a large box may need a simplified version for a label or avatar. Simplifying does not mean shrinking everything until it becomes unclear.

When choosing typography, I would compare real collection names, accented words, prices and product references. A typeface that works in a short headline may have numerals that are difficult to distinguish in an ordering table. I would separate the expressive role of headings from sustained reading in descriptions, while establishing a clear relationship between them.

With files, we need to know what is being delivered. SVG supports vector shapes, images and text. Receiving a file with that extension therefore does not prove that its artwork is constructed as vectors: it may contain a pixel image. I would inspect its construction, curves, single-colour versions and intended sizes. Putting a PNG inside an SVG does not resolve the limitations of the original.

For paper, I would request a proof on the chosen stock. Texture, ink absorption and the printing process can affect the result. A picture of an invented box does not confirm the legibility of a label made by the printer. The decision needs to reach the conditions in which it will be used.

Useful brand guidelines record these decisions and necessary exceptions. They should allow someone else to prepare a new piece without reconstructing the identity from screenshots.

Brand colour cannot decide product behaviour on its own

In its trade ordering area, the same workshop needs to show available products, insufficient quantities and submitted requests. The palette in the presentation may be attractive while offering no answers for these states.

I would separate colour values from their roles. “Dark blue” describes an appearance; “primary text” or “primary action” describes a responsibility. Design tokens, values identified by names, help organise that transition into reusable choices. The U.S. Web Design System explains this use of defined sets of colour, typography and spacing.

In our example, the button for submitting an order should retain a recognisable importance. The link to washing instructions can be quieter, but must remain legible. If both compete with equal emphasis, users have to work out a priority the design should have made clear.

There are concrete checks. At Level AA, the WCAG 2.2 text contrast criterion specifies a minimum of 4.5:1 for normal text and 3:1 for large text, with defined exceptions including logotypes. That exception does not extend to all text because it uses brand colours. Passing this criterion does not certify the accessibility of an interface either.

A quantity error should not rely on a red outline alone. It needs visible information explaining the problem. The criterion on the use of colour requires a visible alternative when colour conveys information. I would place the explanation beside the field and preserve a clear route to correcting the quantity.

Decisions like these connect interface design with identity. The brand retains its character while the application responds to the tasks people need to complete.

The campaign and the film must respect the same intention

For the workshop’s campaign, a generated image might explore a composition or an atmosphere. If it shows a collection being offered for sale, however, I would require it to correspond to the actual pieces. Changing a bowl’s proportions, inventing a glaze or suggesting an unavailable finish changes the promise made to the buyer.

Art direction needs to distinguish exploring a possibility from representing the product being supplied. Good lighting cannot compensate for an inaccurate object. A campaign remains coherent only if its promise carries through to the catalogue, packaging and service.

In a film, intention also becomes a series of choices. For this workshop, I would test shots that reveal the surface and scale of the pieces, allowing time to observe the production gesture. Fast editing might serve a particular advert; it should not be adopted for looking contemporary if it prevents viewers seeing what distinguishes the product.

The same reasoning reaches the ordering area. Photography and language can preserve a sense of closeness while tables remain clear enough for a restaurant to compare quantities. Coherence does not require every medium to use the same composition. It requires knowing what should remain recognisable and what needs to change for each task.

Checking a choice means accepting that it could be wrong

I would start by exposing the system to less favourable conditions: a long product reference, a less polished photograph, an unavailable item and an order containing an error. An identity that works only with content selected for the presentation has not been sufficiently challenged by use.

Next, I would observe people close to the intended audiences. The GOV.UK guidance on moderated usability testing recommends observing specific tasks with instructions that do not reveal the answer. For the restaurant in this example, I would ask someone to find a collection, compare dimensions and prepare an order. I would not tell them where to click.

If someone confuses the finish with the collection name, I would revisit the hierarchy. If they cannot work out how to proceed, I would examine the action and wording before adding more decoration. If they can identify the product but cannot recognise the workshop among similar proposals, the visual direction needs another assessment.

These observations do not justify promising higher sales. They reveal difficulties or support decisions within the conditions observed. Claiming that a solution “converts better” requires evidence suited to that conclusion. My taste does not replace that evidence.

I would also compare an alternative that challenges my preference. If I favour more expressive headings, I want to see them alongside a restrained option using the same content. The question is what each solution makes clearer, more distinctive to the brand or harder to use.

What you should be buying from a professional

Readiness for this work needs to be demonstrated in the proposal and the delivery. The supplier needs to be able to direct the system, verify its applications and recognise what they have yet to master.

That competence needs to be in place before the service is sold. If a supplier can only present variations and cannot assess legibility, application, consistency or suitability, they are not ready to take responsibility for a complete identity. They should limit the service to work they can execute and verify, or involve someone who understands the missing work.

Before approving a branding project, ask to see the connection between the problem and the decisions. Choose difficult applications as well as the most flattering images. Confirm the files, rules and examples you will receive, the uses covered and who responds when a failure emerges within those uses.

That is the work I consider professional: explaining why a solution fits, demonstrating where it works and correcting what verification exposes. The initial image may arrive quickly. Responsibility for the brand begins when someone selects it and presents it as ready to use.