AI image generation has been around for a while. Most of us probably tried it, got a weird 10 finger hand in a photo, and moved on. For a long time, the tools were impressive in some ways but difficult to really use reliably. Especially because text inside images sometimes came out looking garbled. The outputs looked too much like AI generated content to be really useful even if everyone making Studio Ghibli style images took over the internet for a while.
A few things have changed in the last few weeks that make this worth another look. OpenAI just released GPT Image 2 - a rebuilt model that, for the first time, renders readable text inside images reliably. This the kind of thing that makes an infographics, or a flyer look real. You have probably seen these flood your social media feed recently. Last week Anthropic launched Claude Design, a new tool specifically built for layouts like slides, one-pagers, prototypes, and other visuals.
This week's challenge is to experiment with the new image generation models to test them for your specific use case.
Before we start: if you are getting this email forwarded to you and are not yet part of the lab - you can join the tiny AI lab here.
The Challenge:
For this week’s challenge, pick something you want to design. If you are not sure, here are some ideas:
A social media post or infographic
A restaurant menu
A fake movie poster
An invitation
A presentation cover slide
A one-pager or brochure
Unlike for other challenges, I’ve listed some specific tools below since not all AI systems do image generation.
If you are not sure what to use for image generation, ask your AI tool of choice to make recommendations based on the asset you want to create.
For image-based assets: social graphics, posters, quote cards, flyers, movie posters anything where the visual itself is the output. You can try any of the following:
ChatGPT (GPT Image 2): Just rebuilt from scratch and dramatically better at rendering readable text inside images. Free for all users.
Midjourney: Best for high-quality, stylised imagery. Great if aesthetics and visual style matter most to you.
Ideogram: Built specifically for typography inside images. The right choice for posters, social cards, and anything with a headline.
For layout-based assets: slides, one-pagers, menus, invitations, anything where structure and layout matter as much as the visuals.
Claude Design: Anthropic's new tool. Claude builds a first version, you refine through conversation. Export as PDF, PPTX, or send to Canva. Available to Claude Pro subscribers.
Canva AI: Familiar interface with AI generation built in. Good starting point if you already use Canva.
The prompt to get you started
Copy this and run it in your preferred AI system To open it directly in Claude with the prompt pre-loaded, click here or 👉 Open in ChatGPT
"I want to create a [type of asset — social graphic / flyer / menu / poster / presentation slide]. The style should feel [clean and minimal / bold and colourful / warm and editorial / fun and playful]. The main text or message it needs to include is: [your copy or key information]. Generate three variations so I can pick a direction, then I'll tell you what to adjust."
Too Easy?
If the core challenge felt quick or too easy, keep going. Use AI to build a full visual identity for the asset of your choice. The kind of project that would require a designer which would include color pallets, fonts, maybe a logo or a visual assets like a header.
“I just created a [type of asset] for [describe it briefly] (attach image). Now I want to build a visual identity around it. Ask me a few questions about the look and feel I'm going for, then design a complete system: a color palette with hex codes, a font pairing for headers and body text, a logo concept described in enough detail that I could generate it, a one-sentence visual style guideline that ties it all together, and anything else you think could be useful. Once we've agreed on the system, use it to generate a second version of my original asset that applies everything."
Why this matters?
Before AI design tools, if you didn't have a design background your options were limited: learn it yourself, hire someone, or default to generic templates that may or may not fit your brand.
That's changed. When anyone can generate a polished asset in a few minutes, the bar for what's considered acceptable moves up and so do expectations around taste and brand consistency.
It's also reshaping the industry. The World Economic Forum's 2025 Future of Jobs Report, which surveyed 1,000 employers across 55 countries, identified graphic design as the 11th fastest-declining job over the next five years. Two years ago it was categorised as a moderately growing profession. Designers aren't going away but the ones who stay will be those who adapt their offering to what AI can't yet replicate.
This week's experiment is worth running again in six months. The models are improving fast, and tracking where they fall short is the most useful thing you can do to understand the models.
That’s it for this week.
Hit reply and tell me what you thought of the challenge this week.
I read every response, and your experience helps me shape these challenges. If you get stuck or have questions, reach me at [email protected]