AI images for marketing: fast, on-brand and on the right side of the law

What AI images are good for in marketing

Where AI images don’t belong: as proof. A generated photo of a “happy customer”, a team that doesn’t exist or a product result that never happened is misleading. My rule of thumb: trust is a brand’s most important asset. AI images belong where disclosure and quality are right.

Tools compared: pricing and usage rights

Beyond price, the licence matters: can you use the images commercially, and are there conditions for companies? The table shows what vendors’ pages said.

Prices exclude tax, as of Nov 10, 2026 unless stated. Always read the vendor’s current terms before commercial use.

My rule of thumb: use case first, price second. For text in images, Ideogram or ChatGPT often beat pure photo models; for automated workflows an API like FLUX pays off; if you already work in Adobe, Firefly saves editing time.

Prompt anatomy: how to describe an image

Two practical tips: describe colours in words (“sage green”) rather than hex codes, which many models ignore or render as text. And with photo models, add text afterwards in a design tool.

Brand style: making images belong together

One beautiful image is easy. A hundred images that look like they come from the same brand is the real job. Three tools help:

• A style skeleton as a text block. Light, palette, camera, framing, mood and negative list, the same in every prompt.
• Reference images. Many tools accept style or image references. A few approved images as references hold the line better than any description.
• A short image style guide. One page of examples, dos and don’ts, palette and formats that everyone on the team, and every AI, gets.

Editing, upscaling, variants

The first image is rarely the right one. These steps save time:

1. Variants, not a fresh start. If an image is nearly right, generate variants of it rather than rewriting the prompt.
2. Targeted edits. Generative fill or editing by instruction (“replace the mug with a glass”) fixes details without losing the image.
3. Crop to format. Generate larger and crop: 4:3 for web, 4:5 for Instagram, 9:16 for Stories.
4. Upscale carefully. Upscalers invent detail. Never use them unchecked on products, documents or text.
5. Optimise for the web. WebP, several sizes, width and height set. A header image over 300 KB slows the page.

Copyright, trademarks and people

Who owns an AI image?

In its January 2025 report, the US Copyright Office concluded that merely providing prompts isn’t enough for copyright; human-authored elements and creative arrangement or modification can be protected. In the EU, protection requires the author’s own intellectual creation; German law (§ 2(2) UrhG) protects only “personal intellectual creations”. Practical upshot: a competitor may be able to reuse a purely generated visual. The more you shape it yourself, the stronger your position.

The vendor’s licence

Separately from copyright, the terms of use govern what you may do with images. For example: Midjourney requires companies with over $1M in annual revenue to use Pro or Mega; Adobe calls Firefly outputs safe for commercial use but leaves partner models to the user; with some image and video tools, commercial use only comes with a paid plan.

Trademarks, logos and real people

• Other companies’ trademarks and logos don’t belong in your ad images, not even “accidentally” on a mug in the background. Check generated images for recognisable logos.
• Real people may only be shown with their consent; publicity and personality rights apply to AI recreations too.
• Celebrity lookalikes are a risk even if unintended. Regenerate those faces.
• Labelling: if you reach people in the EU, realistic AI images that could pass as real must be disclosed under the AI Act since August 2, 2026. Meta labels detected AI images with “AI info”.

Alt text: required for every image

Every image on your website needs alt text that describes what’s shown. It helps screen-reader users, search engines and AI assistants that read your page. Instagram generates alt text for photos automatically using object recognition, and you can replace it.

• Describe what’s actually in the image, not what you hoped for.
• No “image of”, no file names, no keyword lists.
• With AI images, no “AI-generated” note is needed in the alt text if the page discloses it elsewhere; the description is what matters.
• Around 125 characters max for simple images.

Formats and sizes by channel

Plenty of AI images fail not on the subject but on the format: the face gets cropped in the feed, the text sits under the app’s buttons, the link preview looks tiny. So generate in the right aspect ratio from the start instead of cropping later.

Common values as of Nov 10, 2026; platforms change their layouts occasionally, so check the current specs before big campaigns.

Checklist before use

• Anatomy: hands, teeth, eyes, bodies passing through furniture. Anything slightly warped jumps out at viewers.
• Skin and faces: does the skin look like plastic? Does the face resemble someone famous? Then regenerate.
• Text in the image: generated lettering is often wrong. Add text afterwards in Canva or a design tool.
• Logos and brands: no third-party logos, no recognisable products from other brands.
• Style: does the image fit your brand’s other images, or does it stand out?
• Alt text and labelling: alt text describes what’s shown; realistic AI people get labelled where required.

My rule of thumb: generate four variants and pick the best rather than tweaking one until it fits. It almost always saves time, and the comparison shows you which parts of the prompt actually work.

Product shots with AI: what’s fine

AI is especially tempting for product photos: cut out the product, drop it into a new scene, done. It saves shoots but hides a trap. The product in the image has to match the real one in shape, colour, size and what’s in the box.

• Scene yes, product no: let the AI invent background, light and setting. Take the product itself from a real photo and leave it unchanged.
• No embellished features: if the image shows more than the product does or contains, that’s misleading under consumer protection law.
• Check scale: AI scenes like to make products look bigger or smaller than they are.
• Accessories: anything in the shot that isn’t in the box goes, or is clearly recognisable as styling.

My rule of thumb: a customer opening the box shouldn’t be surprised. If the image promises more than the product, it costs you reviews and returns, and those are more expensive than any shoot.

12 prompts for AI images

The first prompts go to a chat assistant (style, brief); the image prompts go straight to the image model.

Here are three images that fit our brand: [descriptions]. Derive an image style guide: light, palette in words, camera, framing, mood, typical subjects, negative list. Write it as a text block for image prompts (max 80 words).

[Style block]. Header image for an article about [topic]: a person [age, role] [activity], [background], landscape 4:3. No text, no logo, no resemblance to well-known people.

[Style block]. Five visuals for an Instagram series on [topic], portrait 4:5. Same light and palette, a different everyday situation each. Leave the top third clear for text.

My product: [description, exact colours and shape]. Show it in three settings: [setting 1–3]. The product itself must not change: shape, colour and labelling exactly as described. Natural light, no other brands in frame.

Create a mockup of our [packaging/website on a laptop/app on a phone] in a realistic setting: [setting]. Only our own design [description], no other logos, no recognisable brands.

[Style block]. Six image variants for an ad for [offer]: two with a person, two product-focused, two setting-focused. Formats 1:1 and 4:5. Don’t depict results the product can’t verifiably deliver.

Flat illustration in this style: [line weight, palette in words, shapes]. Subject: [subject]. Solid background, no lettering. Matching this series: [description of existing illustrations].

Portrait poster with the text “[exact text]” in large, clear type at the top, [subject] below. Colours: [in words]. Render the text exactly, with no extra words.

In this image, change only [detail, e.g. replace the mug with a glass of water]. Everything else stays exactly the same: person, light, background, framing.

Here’s my image prompt: [prompt]. The result had these problems: [e.g. too dark, distorted hands, text in the image]. Rewrite it to make these less likely without changing subject or style.

Write alt text (max 125 characters) for this image, descriptive and without “image of”: [description of what’s shown and the page’s purpose].

Check this image description for risks: [description incl. people, brands, logos, text]. Are there recognisable third-party brands, resemblances to real people, misleading depictions or labelling duties? A first pass, not legal advice.

Tools around AI images

“Try” links are affiliate links (advertising): I earn a commission at no extra cost to you.

More tools with pricing: best AI marketing tools. For moving images: AI video for marketing.

The most common mistakes

1. Using AI images as proof. Generated “customers” or “results” are misleading.
2. Not checking the licence. Free plans and revenue thresholds can rule out commercial use.
3. A different style every time. Without a style block, the brand looks random.
4. Missing third-party logos or famous faces. Look closely at every image before publishing.
5. Not checking text in images. Models garble letters; typos in ads cost trust.
6. Uploading huge files. Unoptimised images slow pages down.