Visual communication has always rewarded people who understand not just how to make an image, but how images persuade. That second skill — reading the rhetoric of a picture — was already the differentiator between a decorator and a designer. Generative tools have made it the whole ballgame, because the making part is now available to everyone with a keyboard.
The Skill That Just Got More Valuable
When production was expensive, a great deal of design judgment was spent on execution: could you actually achieve the thing you imagined, within budget, before deadline. That constraint quietly did some of the editorial work for you — you made fewer images, so you thought harder about each one. Remove the constraint and the thinking has to become explicit.
The designers thriving right now are the ones who can articulate why a particular composition, palette, or camera angle produces a specific feeling in a specific audience — because that articulation is exactly what a generative prompt requires. Vague creative intuition produces vague output. The vocabulary of visual rhetoric, which design education has taught for decades, turns out to be the most practical prompting skill there is.
What the Tools Are Actually Good At Now
Honest assessment matters more than hype here. Current image models excel at atmosphere, texture, lighting, and stylistic pastiche. They handle typography inside images far better than a year ago but still cannot be trusted with precise copy — professional workflows keep type as a separate layer. Video models produce coherent short clips with synchronized audio, sufficient for social formats, and struggle with long narrative continuity and complex hand interaction.
The workflow that has emerged in studios is neither pure generation nor pure traditional craft. Designers generate to explore quickly, select ruthlessly, then bring the chosen frame into conventional tools for the precision work — type, layout, colour correction, brand compliance. Access to the models themselves has commoditised: most tools reach them through an AI image and video API that fronts many competing models behind one connection, so the practical question has shifted from “which subscription do I have” to “which model suits this brief.”
Teaching Yourself the Next Thing
For designers deciding where to invest study time, three areas repay effort. First, art direction language: build a working vocabulary for lighting, lens behaviour, composition, and mood, because these are the levers that actually control generative output. Second, curation discipline: generate in volume and cut with the severity of a photo editor reviewing a contact sheet — the median generation is mediocre and the best of twenty is often excellent. Third, ethics and disclosure: know what your jurisdiction and your clients’ platforms require, avoid generating identifiable people without consent, and be able to explain your process when asked.
The Reassuring Part
Every technology that democratised production triggered the same fear among practitioners, and produced the same outcome. Desktop publishing did not end typography; it flooded the world with bad typography and made good typographers more visible. Digital photography did not end photography; it made the photographer’s eye the only scarce input. Generative imagery is running the identical play. The tools got cheap. Judgment did not.
One practical exercise worth doing this month: take a finished piece from your portfolio and write the brief that would produce it — not the client brief, but the visual one. Lighting, lens, palette, mood, composition, the emotional register you were aiming for. Most designers find the exercise uncomfortable, because so much of that decision-making happened intuitively. Making it explicit is what turns a generative tool from a slot machine into an instrument, and it happens to make you better at directing photographers and illustrators too.