AI Posters Don't Have to Be Ugly: A Design Guide
The Problem: A Sea of Sameness
AI-generated posters have become ubiquitous, and with that ubiquity comes a distinct problem: they all look the same. A viral Facebook post and subsequent articles in publications like The Independent have highlighted the scourge of 'identikit' posters—AI-created flyers for village fêtes, beer festivals, and local events that share an uncanny, generic aesthetic. The issue isn't just that these posters are bad; it's that they're all bad in the exact same way.
Stephanie Evergreen, a data visualization and design expert, has also noted this epidemic. In her analysis, she points out that AI tends to use a 'maximalist design aesthetic,' cramming in far too many icons and decorative elements. A human graphic designer would exercise discretion, using icons sparingly to concretize abstract ideas. AI, having scraped the internet, simply recognizes that 'posters use icons' and goes all in, resulting in cluttered, repetitive designs that fail to communicate effectively.
The Experiment: Breaking the Mold
John Hartnup, a developer and blogger, decided to tackle this problem head-on. In a recent blog post, he argues that the solution isn't to abandon AI altogether but to use it more intelligently. He set out to prove that ChatGPT is capable of producing a far broader variety of poster styles than the default 'craft fair' look it tends to generate.
Hartnup gave ChatGPT invented event details for a spring fayre and initially asked for a 'clean, unfussy, bright layout with a bold striking spring-themed graphic,' explicitly avoiding pastel, airbrush, or oil-style art and images of people. The result was, in his words, 'still what I'm trying to avoid.' So, he pushed back, asking for a 'completely different design aesthetic' and treating the first attempt as a 'what not to do.'
The Power of Specific Aesthetics
The second attempt was a revelation. ChatGPT produced a Bauhaus-inspired geometric modernist poster—asymmetrical, bold, and visually striking. Hartnup then asked the AI to name the style it had used, and it responded with a detailed breakdown: Bauhaus / Modernist Poster Design, Geometric Minimalism, and Swiss Style (International Typographic Style) influence. This opened the door to a world of possibilities.
Emboldened, Hartnup asked for a list of styles to choose from. ChatGPT provided a diverse menu of 15 distinct aesthetics, ranging from the safe-but-distinct (Swiss Style, Contemporary Editorial) to the left-field (Brutalist, 90s Rave Flyer, Memphis Design). It even offered a steer, suggesting that Risograph, Cut Paper/Matisse, Brutalist (toned down), Japanese Minimal, and Wayfinding/Signage styles would be the strongest stand-out options for his brief.
Results: From Bauhaus to Punk Fanzine
Hartnup generated posters in several of these styles, and the results are compelling. A Stamp/Letterpress style poster has a bold, inked quality. A Japanese Minimal poster is elegant and calm, with plenty of empty space. The Memphis Design poster is playful and bold, with clashing colors and squiggles. He even got a bit silly, asking for a poster in the style of The Designers Republic, the design agency famous for record covers, and a 'photocopied 1980s punk fanzine with some colour snuck in.'
Perhaps most striking is the 'contemporary 40s poster for a cubist exhibition'—a style that feels both historical and modern, and a far cry from the typical AI-generated flyer. Hartnup acknowledges these posters still have a 'whiff of AI' about them, but they are distinctive and, crucially, not ugly. The key takeaway is that you don't have to settle for the default look.
Why This Matters: The Business Case
This isn't just a matter of aesthetics; it's a business concern. As an article in Westword points out, AI-generated ads for restaurants are often 'disgusting' and 'frightening,' failing to present food in an appealing light. These ads don't stand out, don't get attention, and can actively damage a brand's image. The cost of 'free' AI ads is greater than it seems, potentially alienating customers before they even walk through the door.
Stephanie Evergreen echoes this sentiment, arguing that we must accept that AI doesn't inherently know good design. If you choose to use it, you have to know good design yourself to prompt it to success. She advises learning the fundamentals of design (via YouTube or other free resources) and, eventually, budgeting for a human designer who can bring taste and judgment to your materials.
The Future: Editable AI Designs
Hartnup also hints at a further step: asking AI for posters in editable formats. Claude and Gemini can generate HTML, PNG, and PDF where the text is real text and layers are real layers, allowing for font changes, moving elements, and editing text. This moves beyond static images and offers a more professional workflow.
Inspired by his experiment, Hartnup has created a catalogue of one hundred poster styles, complete with ready-to-paste prompts and example images. This resource is a practical tool for anyone looking to leverage AI without falling into the trap of generic design.
The Takeaway: Prompt With Intent
The lesson is clear: AI-generated posters don't have to be horrible. By moving beyond vague prompts and specifying distinct design aesthetics—Bauhaus, Swiss Style, Risograph, Japanese Minimal, Memphis, or even punk fanzine—you can create posters that stand out. The tools are capable of far more than the default 'craft fair' look; it's up to us to ask for it.
However, this requires a baseline understanding of design principles. As Evergreen notes, you must be able to discriminate between good and godawful designs, whether they come from an AI or a human. AI can be a powerful tool, but it's not a replacement for human taste and judgment. Use it wisely, and your posters might just be the ones people remember—for the right reasons.
Related News

Brain Is Two Separate Organs: Stanford Study Overturns Decades of Neuroscience

LLM Writing Guide: Rules, Tools, and AI's Expanding Role

Claude-Assisted Exploit Chain Compromises OpenAI Internal Repos

Bend Language: Prove Code Correct, Run Fast on GPU

4B Model Beats Postgres Query Optimizer by 81%

