Generative AI

What Generative AI Can and Cannot Do for a Design Team

For founders and marketing heads weighing image generators: which design jobs they speed up, which still need a designer, and who should approve the output.

5 min readBy BizHues Engineering

Working on this?

Talk it through with an engineer.

A 30-minute requirements call, no charge.

Book a call

A designer at a jewellery brand in Surat has forty generated Diwali banners open in one window and the brand guide in the other. The founder walks past, sees how fast they appeared, and asks whether the brand still needs two designers. It is a fair question, and the answer depends on which part of design work you mean. Tools such as Midjourney, DALL·E and Adobe Firefly produce visual options quickly; none of them takes responsibility for whether an option fits the brand, the product or the person who will see it on a phone.

Production jobs a generator does well

Four kinds of work suit these tools, and all four are about making material rather than deciding what to use.

Job What the generator gives you What the designer still does
Early exploration Forty directions in an hour Throws away thirty-five, refines the rest
Bulk variation Sizes and formats from one approved idea Checks text, logo and crop in each one
Applying a set style A flat retro look across a set of course cards Writes the style rules the set follows
Simple assets Placeholder or background images Decides whether "good enough" is acceptable here

Bulk variation is where the time goes back to the team. One festive campaign needs a website hero, three marketplace banner sizes, a WhatsApp broadcast image and two test creatives, all following one idea. The designer sets the rules once and checks the results, instead of redoing the same crop twenty times.

In every row, the tool makes the material and a person decides what goes out.

Decisions that stay with a designer

Design is mostly decisions, and decisions need context a prompt does not carry. A model does not know why your brand speaks the way it does, or that customers in Rajkot respond to a different kind of festive imagery than customers in Mumbai. A colour, a symbol or a joke lands differently depending on who is looking, and the generator has no idea who is looking.

Then there is the problem underneath the visuals. If the delivery staff of a dairy cooperative keep abandoning the order-entry screen halfway, fixing it means riding along on a morning route and watching where they get stuck, thumb on a cracked screen, network dropping between villages. No image generator does that. And someone still has to take the sales head's feedback, weigh it against the legal team's, and explain why the final choice is the right one.

Who does what in a design workflow
Generator drafts
  • Concept options
  • Resizes and variations
  • A defined style applied
  • Placeholder images
Designer decides
  • Brand direction
  • Which option and why
  • Usability of the product
  • Stakeholder feedback

The generator drafts material; every decision about what ships stays on the right.

Choosing becomes the main skill

When generating is cheap, attention is what runs short. A designer who used to draw three concepts now has thirty to look at, and the useful skill becomes choosing, combining and explaining. You can only do that well if you know the brand and the audience well enough to say no quickly.

Small physical checks matter more, too. A banner that looks rich on a 27-inch monitor turns into a thumbnail in a WhatsApp broadcast, read on a phone in a bright shop, where the gold text on cream disappears. The person scrolling past is not reading, they are scanning. A designer who knows that rejects most of the forty options before the founder sees them.

For hiring, people who brief clearly, judge quickly and protect the brand become more valuable with these tools. A team without that judgement produces a lot of material that does not hold together, and finds out when the Diwali campaign looks like it came from six different companies.

A person between the generator and the library

We treat the generator as one step in a pipeline that ends with a person approving the asset. The brand and rights check comes before the asset reaches the shared library, not after it is on the homepage.

Generated asset review path
  1. 01Briefaudience, tone, format
  2. 02Generate options
  3. 03Designer curatesdiscard most
  4. 04Brand and rights check
  5. 05Asset librarywith prompt saved

The brand and rights check happens before an asset enters the shared library.

The brief matters as much as the tool. A written brief that names the audience, the format, the tone and what must not appear gives the designer something to judge the output against. Without it, "does this look right" becomes a matter of taste, and taste arguments do not end. The same review idea applies to generated words, which we cover in putting a review step around generated text.

Mistakes that show up after launch

Brand drift comes first. Assets generated by different people on different days look related but not quite the same, and nobody owns the difference until a customer notices that the necklace looks different on Instagram and on the marketplace listing.

Rights come next. Terms of use for generated images vary between tools and change, and teams tend to read them after an image is already on a hoarding. Then reproducibility: nobody saved the prompt, the tool or the settings for the approved image, so when the marketing head wants the same banner with a different pendant, it has to be made again from scratch.

Two smaller ones. Asking for one small change often changes the whole image, so the designer ends up editing the file by hand anyway. And a hundred options help nobody if no one has a free afternoon to look at them properly.

Your first month with generated assets

If you are adding image generation to a small design team, set it up in this order:

  1. Write a one-page brief template: audience, tone, format, what must not appear.
  2. Read each tool's terms for commercial use before anything goes live.
  3. Save the prompt, tool and version next to each approved asset.
  4. Name one person who approves generated assets.
  5. Start with work that is cheap to get wrong, such as internal drafts and placeholders.

When generated imagery moves from marketing into the product itself, the questions change, and our page on generative AI and the post on generated and adaptive interfaces go into those. For this month, name the approver first. Once someone owns what goes out, the other four steps have someone to enforce them.

Share this article
Generative AI

Talk to an engineer

Have a requirement worth building?

Book a 30-minute call with an engineer, at no charge. Tell us the problem and the systems involved, and you leave with questions answered and a suggested first step, not a sales deck.