Image generators have profoundly changed the way visuals are designed. In a matter of seconds, they produce compelling concepts, speed up ideation phases and open up new creative avenues.
But once the initial "wow" effect has passed, another reality emerges: producing an image is not the same as producing a communication asset.
Between a concept and a deliverable ready to be published, there is still a significant gap.
From idea to deliverable, the limits become apparent
For generating a first proposal, the results are often impressive.
In production, the constraints are different. A campaign must respect a visual identity, a layout grid, typography, graphic components and dozens of variations.
It is precisely on these aspects that current tools show their limits.
Faithfully integrating a logo remains difficult. Typography lacks precision. Iterations gradually become inconsistent. Hallucinations still occur: a hand with six fingers, an object that changes shape, illegible text or a detail that seems credible at first glance but does not hold up to scrutiny.
The same observation applies to photographic quality.
At screen size, the result often works very well. When zooming in, materials become artificial, automatic retouching becomes visible and certain details lose all credibility. This level of finish remains insufficient for a poster campaign, a premium brochure or any other medium where the image is viewed up close.
Manual work remains essential
The idea that AI systematically saves time deserves to be qualified.
Creating a first visual is quick. Adapting that visual to different formats, precisely repositioning elements, correcting imperfections or producing dozens of consistent variations still requires a great deal of intervention.
In our practice, some tasks have even become longer. AI generates new variants, but rarely the one that exactly meets the need. As back-and-forth exchanges accumulate, the initial benefit fades.
Today, a graphic designer is often still more effective at finalising a deliverable than a generative model alone.
Changing approach rather than changing tools
At Inox, we have chosen a different approach.
Rather than asking AI to produce images, we ask it to work directly within our design tools.
In practice, a Figma file contains the colours, typography, components and templates defined by our designers according to the client's brand guidelines. An AI agent then controls Figma via its official connector to automatically generate posts, variants or translations from these resources.
The difference is essential.
AI no longer produces a fixed image. It creates an open working file that the designer can review, correct and evolve at any time.
We delegate repetitive tasks without giving up control over the result.
The future belongs to the tools we master
The challenge is probably no longer to replace design software with image generators.
It is about integrating artificial intelligence into the tools teams already use, with their libraries, components and quality standards.
It is this combination that makes it possible to gain productivity without compromising a brand's coherence.
At its core, the question is not whether AI can create a beautiful image.
The real question is who retains control when that image becomes a communication asset.
Today, that is still where the difference is made.
Article author
