Each style page has two ready-made prompts — one for a village fête, one for a music gig. They are starting points, not finished recipes. The general process:

  1. Open ChatGPT and start a new conversation (so previous instructions don’t bleed in)
  2. Copy a prompt and paste it into the chat
  3. Edit the event details — name, date, venue, attractions — to match your actual event
  4. Generate, review, and iterate

Sometimes ChatGPT will respond with text describing what it would do rather than actually generating an image. Before you tell it to get on with it, read what it says — this is a free look at its interpretation of your prompt before any quota is spent. If the description mentions something you don’t want (a colour scheme that sounds wrong, a compositional choice that won’t work, a subject it has misunderstood), now is the best time to correct it:

Actually, make the background dark rather than light. And no figures — just typography and geometric shapes.

Once you’re happy with the direction, follow up with: Do it — or edit and resubmit your original message if you’d rather start fresh.

You don’t need to use the full prompt verbatim. Cut anything that doesn’t fit, or add details that do.

Check your event description before generating. The example events — a village fête and a music gig — are deliberately written with everything a poster reader needs: what, where, when (including the day of the week), and any other key information. Before you generate, ask the AI to review your own event details:

Does this event description include all the information someone would need? Is anything missing or ambiguous?

It’s good at catching omissions: a missing day of the week, a vague venue name, a time that’s implied but not stated.

Sepia, fading, and “antique” effects

Historical styles — Art Nouveau, Art Deco, Constructivist, and so on — can trigger the AI’s instinct to make the image look old: sepia-tinted, faded, foxed, as if photographed from a century-old original. That’s usually not what you want.

If this happens, add one or more of these to your prompt:

vivid, saturated colours — not faded or sepia
clean reproduction quality, as if newly printed
high contrast, no aging or distressing
bold colour, not washed out

For styles where muted tones are intentional (Arts and Crafts, Cubist, some Surrealist work), tone this down — you just want to avoid unintentional degradation, not banish all subtlety.

Inappropriate or awkward imagery

The AI makes things up. If your prompt mentions a band, a town, or a venue, expect it to invent what those look like — realistic-looking photographs of people who don’t exist, plausible-sounding maps of places that aren’t real, street scenes of fictional towns.

A few ways to deal with this:

Make the prompt more abstract. Instead of letting the AI decide what “Carver & the Flood” looks like, specify that you want no photographs of people, no realistic scenes — graphic elements only.

No photographs of people. No realistic crowd scenes. Abstract graphic representation only.

Make the prompt more specific. The AI invents things to fill a vacuum — give it specific instructions and it will do those things instead. Describe the colour scheme, the imagery, the mood, even specific graphic elements you want included.

Deep red and black only. A silhouette of a single guitar. No text other than the event name and date.

Supply a reference image. In ChatGPT you can upload an image alongside your prompt. Upload a photo of your band, your venue, or your town and ask the AI to use it as a reference rather than inventing its own.

Post-edit. Generate the poster as a decorative background and graphic treatment, then add your own text, photographs, and real content on top using Canva, Photoshop, or any basic image editor. The AI-generated part becomes the style layer; your real content sits on top.

These prompts work in other AI image tools too

The examples on this site were generated in June 2026 using ChatGPT (GPT-4o image generation). The prompts should work — with varying results and possibly minor tweaks — in any AI image generation service, including:

  • Google Gemini — usage is more or less identical to ChatGPT; paste the prompt and ask it to generate an image
  • Midjourney — generally strong on aesthetic styles; add --ar 2:3 for portrait format
  • Adobe Firefly — good commercial-use licensing; style adherence varies
  • DALL-E 3 (via API or Bing Image Creator) — the predecessor to GPT-4o image generation; still capable
  • Stable Diffusion / ComfyUI — more control, more setup; these prompts work but may need shorter, more direct wording

Some tools follow detailed prose instructions well; others respond better to comma-separated keywords. If a long prompt isn’t working, try stripping it down to the most important style descriptors and building back up.

Try again — and plan for quotas

No two generations are identical. If your first result is disappointing, try again with the same prompt before changing anything. The variation between attempts is often large enough that a second try produces something much better.

If you’re refining a result, keep your changes small — one adjustment at a time — so you know what’s working.

Look at the result carefully. The examples on this site are mostly the first image that came out of ChatGPT — good enough to illustrate a style, but not necessarily ready to print. For a real poster, zoom in and check: is the text legible and correct? Does anyone have a third arm, a violin bow growing from their elbow, or fingers that don’t add up? AI image generation has characteristic failure modes, and they’re easy to miss at a glance. Reject and regenerate rather than hoping nobody notices.

Think about honesty. Sometimes an illustration is just a pretty picture. But sometimes an illustration makes promises the event can’t keep. If your fête won’t have a big striped marquee with bunting, your poster shouldn’t show one. If your poster shows someone eating candy floss, your event had better sell candy floss. If your event takes place in a community library full of modern furniture, your poster shouldn’t show a Harry Potter oak-panelled library. If your event takes place near a church with a steeple, don’t depict a church with a square belfry. Even if your event does feature a big wheel, don’t depict a forty metre wheel if the real thing will be small.

AI will invent these falsehoods for you if you don’t rein it in.

Ask the AI to review the result too. Before you declare it done, ask:

Is this ready to print and display in the library or corner shop?

It will flag obvious problems — illegible text, a date that got garbled, a layout that wouldn’t survive being scaled up to A3.

Free-tier users: AI image generation quotas are real. Free accounts typically allow a small number of image generations per day. If you’re making a poster for an actual event, don’t leave this to the night before — start a few days ahead in case you need more attempts than your daily allowance allows, or in case you need to wait for your quota to reset.

Paid tiers (ChatGPT Plus, Pro, or equivalents elsewhere) have higher limits but are still not unlimited, particularly for image generation. If you’re generating large numbers of images in one session, pace yourself.

Have your own ideas

One hundred styles is not an exhaustive list. It is a starting point and a demonstration that the space is larger than the default. If you know what you want — a style, a reference, an artist, a decade, a vibe — just ask for it directly. You don’t need this catalogue to tell you that Saul Bass posters exist, or that you liked the look of something you saw last week, or that you want something that feels like a 1970s Soviet children’s book. Describe what you’re after and see what happens.

The prompts here are structured to be reliable and transferable. Your own prompt, written for your specific event with your specific taste, will probably be better.

Consider making it yourself

A human-made poster, even a modest one, is almost always more distinctive than an AI-generated one. AI image generation is good at plausible. It is less good at yours.

If you have any interest in design tools — Canva, Affinity, even PowerPoint — it is worth spending an afternoon on something handmade. You will be able to use your actual photographs, your actual fonts, your actual colours. You will be able to move things around until they look right rather than hoping the next generation is the one. And the result will be unambiguously, unmistakably yours.

This site exists because AI posters are often ugly and repetitive, and they don’t have to be. But the best reason they don’t have to be is that you could just make one.

Did this help?

If these prompts were useful for a real event, I’d love to hear about it — and see the poster you made. Find me on Bluesky.