Image creation with AI: the models and how to use them well

What it is

An image model creates a picture from a written description (a prompt), and many can also edit a picture you give them. Reading an image is a different job (see guide 07).

Think of an illustrator who works in seconds, never asks questions and fills every gap with a guess.

Main tools, as their makers describe them (as of November 2026; check the maker's page):

Why it matters

• Visuals for slides, posters and intranet pages without stock photos.
• Mistakes are easy to miss: misspelt words, extra fingers, wrong logos.
• Pictures of real people, brands or artists' styles carry legal and trust risks.

How to do it

1. Check which image tools your organisation approves, and for what. Never upload confidential material or photos of people unless allowed (see guide 10).
2. Describe five things: subject, style, composition (layout, viewpoint, lighting), any exact text, and aspect ratio (shape, such as square or wide).
3. Edit one change per request rather than starting again.
4. Keep the brand consistent with reference images, a brand kit and a saved team prompt (see guide 11).
5. Review: a person checks every word, hand, face, logo and number before use.
6. Label it AI-generated where people could be misled, and note the tool and prompt.

Example

Before: "A picture for our safety newsletter" gives a generic scene with unreadable signs.

After:

Subject: a tidy office kitchen with a kettle and a floor sign.
Style: flat, friendly illustration, soft colours.
Composition: kettle left, yellow sign centre-right, eye level,
empty space at the top for a headline.
Text on the sign, exactly: "Wet floor". No people, logos or other text.
Aspect ratio: wide, 16:9.

You check the sign, add the headline in your design tool, and caption it "Image created with AI".

Common mistakes

• Generating a real person, colleague or celebrity. Fix: do not, unless they agreed in writing and your policy allows it.
• Copying an artist's style or a competitor's brand. Fix: describe colours, mood and shapes instead.
• Trusting text inside the image. Fix: proofread it, or add text afterwards in a design tool.
• Assuming you own the result. Fix: check the terms with your legal team before commercial use.

Try it this week

1. In an approved tool, write a five-part prompt for a slide illustration with no people or logos (6 minutes).
2. Make two single-change edits, such as colour and layout (5 minutes).
3. Check every detail, label it and save the prompt (4 minutes).

Going further

• Content Credentials (open standard C2PA) attach a record of where a file came from and how it was edited. OpenAI, Adobe Firefly and Microsoft Designer add them.
• SynthID is Google's invisible watermark; Google says all images made with the Gemini API carry it, and OpenAI says its images carry it too. Google's checker also covers video and audio.
• EU AI Act, Article 50 (European Union; general information, not legal advice): from 2 August 2026, deepfake images used at work must be disclosed; see guide 44.
• Screenshots and heavy edits can remove these marks, so their absence proves nothing.
• Indemnity usually covers only paid business plans, unmodified output and correct use.

Key words

• Image model: an AI that creates or edits pictures from a written description.
• Aspect ratio: an image's shape, such as 1:1 (square) or 16:9 (wide).
• Content Credentials (C2PA): attached information recording how a file was made and changed.
• Watermark: a visible or hidden mark showing an image was AI-generated.
• Indemnity: a maker's promise to defend you against certain legal claims, under conditions.

Sources

Examples and figures in this guide are illustrative. AI tools change often: check what your organisation's approved tools offer today. Use approved tools only, and a person checks and decides.

Prefer to print or share it? Download this guide as a PDF (296 KB), or the complete AI Field Guide (2.5 MB).