AI image generation: from a sentence to a picture
Image generation is an AI's ability to create a brand-new picture from a short written description. If you have ever asked an assistant to « draw me… », you have already used it.
Data as of October 22, 2026
Key figures
| Level | Beginner |
|---|---|
| Origin | 2014 |
| Introduced by | Goodfellow et al. (réseaux antagonistes génératifs, GAN) |
How a sentence turns into a picture
You type an instruction, called a prompt: « a ginger cat reading the newspaper on a city balcony, watercolor style ». A few seconds later, the tool returns an image that didn’t exist before. It’s one of the most visible uses of generative AI.
Behind the scenes, a model was trained on a huge collection of images paired with descriptions. It learned the links between words and shapes, colors and styles. A key milestone came in 2014, with the founding paper on generative adversarial networks (GANs) by Ian Goodfellow and co-authors. Today, most consumer image generators rely on diffusion models, which start from random « noise » and refine it step by step into a coherent picture.
Everyday uses
- At work: illustrating a slide deck, creating a header image for a blog post, exploring several visual directions before briefing a designer.
- At home: picturing a room makeover, making a personalized birthday card, illustrating a bedtime story.
- Photo editing: many tools can also modify an existing photo, such as swapping a background, extending a scene or removing an object.
Several mainstream assistants, such as ChatGPT and Gemini, offer this feature right inside the chat.
Why it matters to you
In practice, the quality of the result depends mostly on how precisely you describe it: subject, style, framing, mood, colors. The prompt generator can help you structure that request.
That ease of use has a downside: a realistic image can also be used to deceive. Deepfakes rely on the same techniques. That’s why some AI makers embed an invisible watermark in their images, and EU rules now require certain generated content to be disclosed.
Common misconceptions
- « The image is automatically copyright-free. » Not necessarily: the rules depend on the tool and the country, and an image imitating a brand or a real person can cause problems.
- « It’s searching a stock photo library. » No, the picture is created on each request; two identical prompts often give two different results.
- « The first try is the right one. » Rarely: people usually refine the prompt over several rounds.
- « The AI understands what it draws. » It produces likely shapes, which explains mistakes with hands, reflections or text inside the image.
Related terms
- Generative AI · AI that creates new content such as text, images, audio, video or code from a simple written or spoken request.
- Diffusion model · A type of generative model that creates an image (or audio, or video) by starting from random noise and cleaning it up step by step.
- Prompt · The text, question or instruction you give an AI to get a response; how clear it is directly shapes the result.
- Deepfake · An AI-made or AI-altered image, video or voice recording that makes a real person appear to say or do something they never did.
- AI watermarking · A mark, often invisible, embedded in an image, audio, video or text to signal that it was generated by AI.
- Multimodal AI · AI that can understand or produce several kinds of content, such as text, images, audio and video, often within the same conversation.
Try it
Full glossary
Frequently asked questions
Does the AI copy existing images?
As a rule, it doesn't paste an existing picture: it produces a new one based on what it learned during training. It can, however, produce something very close to a famous artwork or a protected character, especially if you ask for it. Be careful with any public or commercial use.
Can I use AI-generated images on my website or social media?
It depends on the terms of use of the tool you used, and they differ from one service to another. Read them before any commercial use. Also avoid depicting real people without their consent.
Why are hands or text inside the image sometimes wrong?
The model reproduces likely shapes without truly understanding what a hand or a word is. Recent models have improved a lot, but always check details such as fingers, lettering and logos before publishing.
How can I tell whether an image was made by AI?
Some tools add an invisible watermark or provenance metadata. This isn't universal, and such marks can disappear after certain edits. A careful eye for inconsistent details is still useful.
Data sources
Data as of October 22, 2026
Report an error
Does a figure, a date or a fact look wrong? Tell us: every report is checked.