Creating images with AI: how to get exactly the picture you want

What does creating images with AI mean?

You write in plain language what kind of picture you want, and the AI draws it. You might ask for "a summer cottage jetty in evening light, in watercolour style" and get a brand-new image in seconds. You can then change it simply by continuing the conversation: "add a red rowing boat by the jetty" or "make the colours warmer".

The AI does not fetch the picture from the internet. It has learned from a huge number of images what things usually look like, and it creates a new picture based on that. This is why the result is often surprisingly good, but sometimes odd details slip in, such as six fingers or letters that do not form real words.

Which tool is good for what?

There are plenty of good options, and most of them let you start for free. Free versions usually limit how many images you can make per day or month.

• ChatGPT – the easiest place to start if you already use it. Ask for a picture right in the chat and refine it by saying what to change. It handles text in images fairly well, too.
• Gemini – Google's AI, which is especially good at editing an existing picture: upload your own photo and ask it to change the background or add something.
• Canva – best when the image needs to become a finished card, poster or social media post. The AI picture lands straight in a template with text and layout ready.
• Adobe Firefly – Adobe's tool, trained on licensed material. That makes it a safer choice for businesses using images commercially.
• Midjourney – known for artistic, atmospheric images. It is a paid service and takes a little more learning.
• Ideogram – particularly good when the picture must contain readable text, such as a poster or a sign.

How do you describe a picture so it turns out well?

The most important skill is describing. The AI cannot read your mind, so tell it the same things you would tell a photographer or illustrator. A good description answers five questions:

1. What is in the picture? The main thing first: "an old red wooden cottage".
2. Where and when? "on a lakeshore, on a summer evening at sunset".
3. In what style? "photograph", "watercolour", "children's book illustration", "simple line drawing".
4. What mood and colours? "peaceful, warm orange tones".
5. Where will it be used? "landscape", "square for Instagram", "portrait for the front of a card".

Do not try to get a perfect picture in one go. Ask for a first draft and then adjust in small steps: "fewer clouds", "move the cottage closer", "remove the people". You can learn more in the guide How to talk to AI.

Example 1: a birthday card

A greeting card is a great first experiment. Make the picture with AI and preferably add the words yourself, for example in Canva, because AI can misspell names and greetings.

When the picture is ready, open Canva, choose a card template and add the picture and your own greeting. That way the name and age are certain to be right.

Example 2: a product photo for a small business

If you sell crafts, baking or other products online, AI can help make your photos more appealing. The best approach is to use your own photo as a starting point: photograph the product in even light and ask the AI to change the background.

Always check that the product looks the same in the final picture as it does in real life. If the AI changes the colour, pattern or size, customers will be disappointed. You will find more ideas in the guide Marketing and social media.

Logos: a good sketch, a poor final result

Many business owners want to make their logo with AI, and it is excellent for finding ideas. As a finished logo, however, an AI image is often a weak choice:

• An AI image is a pixel image, while a logo usually needs to be in vector format so it works just as well on a business card as on the side of a van.
• The AI may accidentally produce something that resembles an existing logo.
• An image made solely by AI may not be protected by copyright, which makes it hard to protect against competitors.

A good approach: create ten ideas with AI, pick a direction, and then have a graphic designer, or Canva's design tools, help you draw a clean logo of your own.

Copyright and real people

You may edit pictures of yourself and your family when those involved have agreed to it. Be open about it if you publish an AI-made picture, for example in an advert. You can read more in the guide AI and the law, and the guide AI scams explains how fakes are used in fraud.

When the picture is not right: common problems

• The picture is too ordinary. Add details: season, light, materials, camera angle. "Shot from a low angle", "close-up" or "aerial view" change the mood a lot.
• Hands or faces look strange. Ask for a new version or crop the picture so the hands are not visible. You can also ask for the people to be shown from further away.
• The AI changes too much. Say clearly what must stay the same: "change nothing except the background colour".
• The style varies from picture to picture. When you find a style you like, save the description and reuse it for the next picture. That way your posts look like they belong together.

Remember too that AI does not always understand "no". If you write "no dog", a dog may still appear. Describe what is in the picture rather than what should not be there.

How to get started today

Open ChatGPT or Gemini and ask for a picture of a place that is dear to you: your childhood home, the shore by your summer cottage or your favourite café. Describe it using the five questions and adjust the result two or three times. One evening of experimenting will teach you more than any set of instructions. When you want to give your pictures movement or sound, continue to the guide Video and audio with AI.