How to Generate Images with AI: The Complete Guide

To generate images with AI today, you simply describe what you want to see in a few sentences, either in an assistant such as Gemini, ChatGPT, Copilot or Le Chat, or in a dedicated tool such as Adobe Firefly. This guide explains how these tools work, which one to choose, how to write a good prompt, how to edit the result and how to use it while respecting other people’s rights.

How an AI creates a picture

Image generation means producing a brand-new picture from a written description, sometimes together with a starting photo. The tool does not paste together pieces of existing images: the model creates every pixel.

Most tools rely on a diffusion model. Put simply, during training the model learns to gradually remove “noise”, a kind of random static, from images paired with their descriptions. When it creates, it starts from pure static and cleans it up step by step, guided by your text, until a coherent picture appears. Other approaches exist, and companies do not always disclose the exact architecture of their models.

Recent assistants are also multimodal: they understand both text and images. That is why you can upload a photo and ask “add a sunset sky” or “turn this sketch into a watercolor”. For the bigger picture, our explainer on generative AI covers the basics.

Two practical consequences to keep in mind:

• Results change every time. The same prompt produces different pictures. That is normal, and useful: generate several versions and keep the best one.
• The model does not “know” what is true. It produces a plausible image, not a photo of a real event. A six-fingered hand, garbled lettering or an impossible object can still appear, even if recent models make fewer of these mistakes.

The tools for generating images in 2026

There are two families: general-purpose assistants you may already have on your phone, and dedicated creative tools.

Assistants you probably already use

According to the profiles on our site, compiled from official pages as of October 22, 2026, most major assistants can create images:

• Gemini (Google): the free version includes image generation and editing. According to Google’s official image generation page (accessed October 22, 2026), the free model is called Nano Banana 2, and paid Google AI plans unlock Nano Banana 2.1. Google highlights changing the mood or lighting, applying the style of another photo, and readable text in many languages for posters, invitations and comics.
• ChatGPT (OpenAI): creates images on every plan, with limits on the free version; the Go plan raises the number of images you can make.
• Microsoft Copilot: image creation is available without a subscription, “subject to available capacity” according to Microsoft.
• Le Chat, now called Vibe (Mistral AI): generates images, with “basic images” on the free plan.
• Meta AI: creates images right inside WhatsApp, Instagram, Facebook and Messenger; Meta One subscriptions give more room to create.
• Grok: generates images and short videos with its Imagine feature.
• Qwen Chat (Alibaba): image generation and editing, in a service its publisher describes as free.
• Apple Intelligence: Image Playground and Genmoji on compatible Apple devices.

By contrast, Perplexity’s free version does not include image generation, according to its profile.

Assistant	Free	Paid from	Web search	Agent mode	Voice	Images
ChatGPT	Limited	$8/mo	Yes	Yes	Yes	Yes
Gemini	Limited	€4.99/mo	Yes	—	Yes	Yes
Microsoft Copilot	Limited	$9.99/mo	Yes	—	Yes	Yes
Le Chat (Mistral Vibe)	Limited	$5.99/mo	Yes	Yes	Yes	Yes
Meta AI	Limited	$7.99/mo	—	—	Yes	Yes
Grok	Limited	$8/mo	Yes	Yes	Yes	Yes

An empty cell in this table means the company does not state the information, not that the feature is missing. To weigh these criteria against your other needs, the AI assistant comparison tool lets you filter by feature.

Dedicated creative tools

Specialized tools offer more controls: fine editing, upscaling, batch work and project management. Here is what their makers state on their official pages.

Tool	Free access according to the company	What the company highlights
Adobe Firefly	Yes, a limited number of free daily generations	Images, video, audio, editing (removal, expansion, upscaling), models from Adobe and partners
Runway	Free trial	A platform to generate and edit video, images and audio
Genspark	Yes, free plan with credits	Workspace with agents, including image and video generation
Poe	Yes, daily points	Image generators from several providers, priced in points per image

Sources: Adobe Firefly page and Runway website, accessed October 22, 2026; our site profiles for Genspark and Poe.

According to Adobe, Firefly gives you access to its own models and to partner models from companies such as Google, OpenAI, Runway and Kling AI, with a single login. Adobe says its Firefly models are trained on licensed Adobe Stock content and public domain works, which matters to professionals who care about rights.

Writing a good image prompt

A prompt is the instruction you give the tool. For an image, a vague phrase like “a cat” gives a generic result. A good prompt spells out six elements:

1. The subject: who or what, with a few details (“a ginger cat sitting on a windowsill”).
2. The action or situation: what is happening (“watching the rain fall”).
3. The setting: place, era, season (“in a London flat, in autumn”).
4. The style: realistic photo, watercolor, flat illustration, pencil drawing, 3D render.
5. Lighting and framing: soft morning light, backlighting, close-up, top-down view.
6. The format: square for social media, landscape for a banner, portrait for a poster.

A complete example: “Watercolor illustration of a ginger cat sitting on a windowsill, watching the rain fall over London rooftops in autumn, soft grey light, horizontal framing.”

If you want text in the image, for an invitation or a poster, put the exact words in quotation marks in your prompt and check every letter. This keeps improving, but mistakes still happen.

Then refine with small changes instead of rewriting everything: “same image, but with a brighter sky”, “keep the cat, move the scene to a kitchen”. Our prompt generator helps you build a complete prompt when you run out of ideas.

Check yourself You want an image for a child's birthday invitation. What do you add to “a birthday cake”?

The style (a colorful, cheerful illustration), the setting (a party table with balloons), the format (portrait, for a card), and the exact text to display in quotation marks, such as the child’s name and age. Then you check the generated text letter by letter.

Editing and improving an image

Generation is only the first step. Tools increasingly offer editing features, as described by their makers:

• Changing part of the image: swap an object, a color or the background by describing it.
• Changing the mood: Google mentions switching from daylight to nighttime lighting, or changing the viewing angle.
• Applying a style: reuse the texture, color or style of another photo, according to Google.
• Expanding or upscaling: Adobe highlights removing elements, extending the frame and increasing resolution.

One precaution: when you upload a personal photo for editing, it goes to the company’s servers. Avoid photos of other people without their consent, and photos that show documents or private information.

Everyday uses

Image generation is useful well beyond having fun. Some common uses:

• Illustrating a document or slide deck: a mood image for a meeting presentation or a school project. Our tutorial on how to create a presentation with AI shows how to fit it in.
• Making an invitation or a flyer: a party, a yard sale, a community event.
• Visualizing a project: a garden layout, a wall color, a logo idea to hand over to a designer afterwards.
• Animating an image: several tools turn a still picture into a short clip. Everything is explained in our guide to creating videos with AI.
• Adding sound: background music, a voice-over. See our guide to AI-generated music and voices.

Rights, transparency and good conduct

Creating an image in seconds does not exempt you from a few rules.

Copyright. Terms of use vary from one tool to another, and the legal status of generated images is not the same everywhere. Avoid imitating a named living artist, reproducing a protected character or a brand. Our article on AI and copyright sets out the main principles as general information. For a commercial project, consult a legal professional.

Real people. Creating a realistic image of an actual person in a situation that never happened can mislead, hurt, or amount to a deepfake. Most companies restrict or ban this in their terms.

Transparency. Several companies mark their images. That is the role of an AI content watermark.

These marks help, but they do not cover every tool, and metadata can disappear with a simple screenshot. To learn how to spot an artificial image with no mark, read our guide on how to recognize an AI-generated image.

Where to start

Here is a simple four-step method for your first image:

1. Open an assistant you already have, without paying.
2. Describe your image using the six elements above.
3. Generate several versions, then refine with small changes.
4. Before publishing, check the text in the image, the tool’s terms of use, and that no recognizable real person appears.