You need a visual for a presentation or a post. An AI can produce one in seconds. But between "I type a sentence" and "I have an image I can actually use," there are a few steps: choosing your tool, making a clean first attempt, avoiding the pitfalls, and above all checking your visual before you show it.

And if the visual is meant to illustrate slides, start with the message and the outline: our guide to making a presentation with AI covers the full method.

Step 1: choose a type of tool

No need to compare fifty generators. There are three main types.

The generator built into an assistant

Several assistants, such as ChatGPT, Gemini, Copilot, or Le Chat by Mistral, can create images right in the conversation (available features depend on the plan and change over time). You describe, you get an image, you ask for changes in plain language. The easiest place to start.

The dedicated image tool

Services like Midjourney are designed just for images. More settings (format, variations, consistency across a series), more control over the result, and a bit more of a learning curve.

The tool built into design software

Some layout or photo-editing programs generate images directly in your document. You then add your text and logo with real design tools.

The advice: start with the assistant you already use. You'll switch tools the day you know exactly what you're missing.

Step 2: your first attempt, step by step

  1. Write down the use before the prompt. Note for yourself: "horizontal banner for a post about the community group's cooking workshop." How you'll use the image determines the format and the level of detail.
  2. Describe what you see, not what you feel. Objects, a place, a type of light. To structure your description, the 5-layer grid does the job.
  3. Specify the format. Square, landscape, portrait. A square image cropped into a banner often loses what matters most.
  4. Generate several options. Don't judge the tool on a single image.
  5. Pick the best starting point, then refine. Not the perfect one: the closest one.

An example first attempt:

Generate a horizontal image to illustrate a community cooking workshop. A large wooden table with seasonal vegetables, hands slicing tomatoes, colorful aprons. Realistic photo, natural late-morning light. No text in the image.

Notice the last sentence: we'll come back to it below.

Why two attempts give two different images

You run the same request again and get a completely different image. That's not a bug. A generator starts from random noise and turns it, step by step, into an image that matches your description. Each attempt starts from a different point, so it ends somewhere different too. Your description guides the result; it doesn't lock it in.

Two practical consequences:

  • Save the images you like as soon as you see them. You won't necessarily get them again.
  • If a specific detail matters, write it down. Anything you don't specify is left to chance and will change from one attempt to the next.

The classic pitfalls

Hands

Extra or fused fingers, impossible wrists: this is still the first place to look. If the hands aren't essential, frame the shot so they're barely visible.

Text in the image

Distorted letters, made-up words, typos. The safest approach: ask for an image with no text, then add your title in a layout tool.

Logos and brands

Avoid asking for a real brand's logo or a recognizable product: the result is often wrong, and it raises trademark issues. For your own logo, add the official file afterward.

And if you actually need to come up with that logo, follow the dedicated method to create a logo with AI, from the brief to a clean file.

Faces of real people

Generating a real person raises questions about image rights and deception. Many tools refuse to do it. For a professional visual, stick to fictional characters.

A living artist's style

Asking for something "in the style of" a living illustrator means using their work without their consent. Describe the style instead: "thick ink lines, pastel colors, flat color with no gradients."

Rights and terms of use

Commercial use, image ownership, prohibited content: every tool has its own rules, and they change. Before any professional use, read the terms of use of the tool you're using.

The checklist before using a visual

You proofread a generated image just like a text. Before publishing it, go through these six points:

  • Zoom in on the details. Hands, eyes, jewelry, chair legs. Look at it full size, not as a thumbnail.
  • Text. Every letter, including writing in the background (signs, labels, screens).
  • Light and shadows. Do all the shadows fall on the same side? Do the reflections match the light source?
  • Extra elements. A third arm, a floating object, a cup melting into the table.
  • The real message. Show the image to someone without explaining anything. If they don't get your message, the image is pretty but it failed.
  • "AI-generated image" label. If it could be mistaken for a real photo (an event, a place, a person), say so.

Edit instead of regenerating everything

Your image is almost right, but there's a flaw. Don't start over: you'd lose what works and go back to chance. Edit instead:

  • Editing a specific area. Many tools let you select part of the image and regenerate only that part.
  • A targeted request. In an assistant, describe only the change:

Keep exactly the same image, same colors, same framing. Only replace the blue cup in the foreground with a white cup.

  • Classic photo editing. For a tiny detail, traditional editing software is often faster.
  • One change at a time. If you change three things at once, you won't know which one broke the rest.

These habits also apply when you make a video with AI, where every attempt costs even more time.

Going further

To understand what happens between your sentence and the image, play the free chapter of the Prism island, "From Noise to Image," in your browser with no account: start the chapter. Then test yourself with the friend challenge: ten timed questions about AI, and a score to beat.

If your image is meant for a post, our guide on how to write social media posts with AI covers the text that goes with it, and how to stay honest about what is real and what is generated.

Frequently asked questions

Which tool should a beginner use to generate an image with AI?

The easiest is to start with the assistant you already use, if it can generate images in the conversation. You'll move to a dedicated tool when you need more settings or more control over the result.

Why does the same request give a different image every time?

The generator starts from different random noise on each attempt, then turns it into an image based on your description. Your request guides the result without locking it in, so anything you don't specify changes from one image to the next.

Can I use an AI-generated image for commercial purposes?

It depends on the tool: each service sets its own terms of use, and they change over time. Read the terms of the tool you're using before any commercial use, and avoid logos, brands, faces of real people, and the styles of living artists.

Do I need to say that an image was generated by AI?

Yes, whenever someone could mistake it for a real photo, for example of an event, a place, or a person. For a clearly fictional illustration, labeling it is still good practice for transparency.