How to Use AI Image Generation for Small Business Marketing

Every small business needs pictures. Social posts, a website refresh, a flyer for the shop window, product shots, a header for the newsletter nobody has sent since March. Until recently the options were a stock-photo subscription that made you look like every other business on the internet, a photographer you could not justify, or your phone and some optimism.

AI image generation changes that maths. For roughly €20 a month you can produce usable marketing visuals in minutes. But most owners have the same complaint about AI image advice as about AI advice generally: “every article says use AI but nobody says how.” So this is the how — the tools, the prompt structure that works, the brand-consistency problem nobody warns you about, and the legal questions you cannot skip.

What AI images are actually good for — and what they are not

Getting this wrong is what makes AI imagery look cheap. AI generation is excellent at illustrative and atmospheric visuals and poor at depicting your actual business.

Things it does well today:

• Abstract and conceptual images — a blog header about cash flow, a slide background, an email banner. Nobody expects these to be real.
• Illustrations and icons in a consistent style: line drawings for a how-it-works section, simple spot illustrations for a pricing page.
• Mock-ups and scene composition — putting your real product photo onto a plausible table, wall or shelf, or showing a packaging concept before you commit to a print run.
• Variations at speed. Nine versions of an ad background in four aspect ratios takes about three minutes — the real unlock for anyone running paid social.

Things it does badly, or should not be doing at all:

• Your team, your premises, your food, your finished work. A generated “happy customer” in your reception is a fabrication, and customers are getting good at spotting them.
• Accurate detail. Hands, tools, machinery, text on signage, the number of legs on a chair. Models have improved; they are still unreliable on specifics.
• Before-and-after proof. If you are a builder, a cleaner, a dentist or a landscaper, your before-and-afters are your most valuable asset. Generating them is fraud, not marketing.

The rule of thumb: if a customer would reasonably assume the photo is real evidence, do not generate it. Use AI for mood, metaphor and layout. Use a camera for proof.

The tools worth paying for in 2026

“There is a new AI tool every week” is the most common sentence in our inbox, so here is a deliberately short list. You need one generator and one editor. That is it.

Already in a subscription you probably pay for. ChatGPT Plus and Google Gemini both include image generation at no extra cost, and Microsoft Copilot includes it on Microsoft 365 business plans. For most SMBs this is genuinely enough. Before you buy anything, spend a fortnight generating with the tool you are already paying for — it answers the “which one is actually worth my money?” question for free.

If image quality is the product. Midjourney (around €10–€27 a month depending on tier) still produces the most attractive output and the best style consistency, which matters if visuals are central to your brand. The trade-off is a steeper learning curve and a workflow that lives outside your other tools.

If you need brand assets, not just pictures. Canva’s paid tier (around €12 per user per month) bundles generation with templates, brand kits, background removal and resizing. For a one-person marketing function it is usually the highest-value single purchase: generating the image is the easy part, getting it into eleven correctly sized formats is the slow part.

If provenance matters to your clients. Adobe Firefly, included with Creative Cloud, is trained on licensed and public-domain material — worth it for agencies with cautious clients.

Spend under €40 a month in total. If you are above that on image tools alone, you have a stack problem, not a capability problem — our guide to the best AI tools for small business in 2026 covers how the pieces should fit together.

How to write an image prompt that does not look generic

Most disappointing AI images come from prompts of four or five words. The fix is a repeatable structure. Write prompts in this order:

Subject → action → setting → lighting → composition → style → what to exclude.

Compare these two. The first is what most people type:

A plumber at work

The second takes fifteen seconds longer and is usable:

A plumber in a navy work shirt tightening a valve under a kitchen sink, small British terraced-house kitchen, warm daylight from a window on the left, shot from low at the subject’s eye level, shallow depth of field, documentary photography, muted colours, no visible brand logos, no text

Four things do the heavy lifting. Specific clothing and setting stop the model defaulting to glossy American stock photography. Named lighting is the single biggest quality lever. Camera language — angle, depth of field — makes output read as photography rather than render. And explicit exclusions save you re-rolls: “no text, no logos, no extra hands.”

Three prompts worth saving as templates today:

Blog header. “Wide 16:6 banner illustration representing [topic], flat vector style, two-colour palette of deep blue #0D47A1 and warm orange #E65100 on white, generous negative space on the right third for headline text, no words or letters in the image.”

Product in a scene. “[Your product] placed on a pale oak table beside a linen napkin and a ceramic cup, soft diffused morning light, top-down composition, square crop, Scandinavian editorial style, neutral tones, no people, no text.” Then composite your real product photo into the generated scene rather than letting the model invent your product.

Keep your winners in a shared document with the tool and settings that produced them. A prompt library is a business asset; a chat history is not. The same discipline that makes AI-assisted social media management work applies here.

Keeping your brand consistent across generated images

This is where AI imagery most often falls apart for small businesses. Twenty individually attractive images in twenty different styles look worse than twenty mediocre images that match. Four practical controls:

1. Write a one-page image brief and reuse it verbatim. Two or three sentences of fixed visual rules — palette in hex, illustration or photography, warm or cool, people or no people. Paste it into every prompt. Ours is: deep blue and orange on white, generous white space, no stock-photo smiling.
2. Use style references, not just words. Midjourney’s style reference, ChatGPT’s uploaded examples and Canva’s brand kit all anchor new images to ones you have already approved — far more reliable than describing a style in prose.
3. Fix the aspect ratios up front. Decide your five formats — 16:9 for blog headers, 1:1 and 4:5 for social, 9:16 for stories, 1200×630 for link previews — and generate into them rather than cropping afterwards.
4. Compress before publishing. Generated images often arrive at 2–5 MB. Resize to the width you display and export as WebP — a 300 KB hero loads, a 4 MB one costs you the visitor.

The legal and trust questions you cannot skip

Four things to settle before this becomes a habit across your team. All are cheaper to decide now than to unpick later.

Who owns the output. Most major tools grant commercial-use rights on paid plans, and some free tiers do not. Copyright in purely AI-generated work is also weaker than you might assume in the UK, the EU and the US, which matters if you want to defend a logo or a character. Check your plan’s terms, keep a note of which tool produced which asset, and read our breakdown of who owns AI-generated content before you build anything load-bearing on a generated image.

Never generate a real person. Not a celebrity, not a competitor, not a customer, not a staff member who has not agreed in writing. That is the fastest route to a legal letter.

Logos and recognisable styles. Do not prompt for another company’s branding, and avoid “in the style of [living artist].” Describe qualities — flat, geometric, hand-drawn — rather than borrowing a name.

Disclosure. Under the EU AI Act, certain AI-generated or manipulated content aimed at the public carries transparency obligations, and platform rules are tightening in parallel. Nobody expects a label on abstract or illustrative imagery. Label anything a viewer could mistake for a photograph of real people or events. Many tools now embed C2PA content credentials automatically — leave that metadata intact.

Write these four decisions into your AI policy as a short section on imagery, with one named owner. A rule that lives only in your head is not a policy.

A 30-day rollout for a small marketing team

You do not need a project plan. You need four weeks of small, specific moves. “No step-by-step examples for my actual business” is the complaint; here is the step-by-step.

Week 1 — decide and brief. Pick the generator you already pay for. Write the one-page image brief: palette in hex, style, people or no people, the five aspect ratios. Agree three red lines — no real people, no fabricated proof, no competitor branding.

Week 2 — build the prompt library. Choose your three highest-volume image jobs. Generate until you have one prompt per job that works twice in a row, then save prompt, tool and settings in a shared document. Expect to discard 80% of what you generate.

Week 3 — replace one real workflow. Take the next month of social posts or blog headers and produce the images from your library. Time it honestly. If a header used to take 25 minutes of stock-photo hunting and now takes five, you have saved roughly three hours a month — a defensible number for the subscription.

Week 4 — review and write it down. Look at the month’s output side by side. Does it look like one brand? Cut what does not fit, note the prompts behind the keepers, and add the imagery section to your AI policy.

Measure one thing: images produced per hour of human time, at an acceptable quality bar. If it has not moved by week four, the problem is your prompt library, not the tool.

The bottom line

AI image generation is one of the few AI capabilities that pays back almost immediately, because the alternative — stock subscriptions, freelance design, or going without — was already costing you money or credibility. The owners who get it right are not the ones with the most expensive tool. They are the ones who decided what their brand looks like, saved five prompts that produce it, and drew a clear line between images that set a mood and images that make a claim. Keep that line and generated visuals quietly improve your marketing. Cross it and one fabricated photo costs you more trust than a year of good posts earned.