With ChatGPT image generation, you describe what you want to see in the text box. ChatGPT then generates an image based on your description within seconds. The feature works in the web version, the desktop app and the mobile apps.
You do not need any design skills. All that counts is a clear description. The more specific your prompt, the more closely the result matches what you have in mind. All users have access to the image feature, including those with a free account.
How does ChatGPT image generation work?
ChatGPT’s image feature turns text into images through an underlying image model. The current image model is called ChatGPT Images: version 2.0 became available on all ChatGPT plans on 21 April 2026, and on 8 September 2026 ChatGPT Images 2.5 followed, with sharper details, more precise edits and faster image generation.
Technically, ChatGPT does not retrieve existing images. It creates a new image based on patterns the model learned during training. Every image is new, even if you enter the same prompt again. The model combines your description with learned visual patterns and produces a result that matches your instruction as closely as possible.
Ever since GPT-4o’s native image generation (GPT Image 1, March 2025), the successor to DALL-E 3, the image model has been better at handling text in images, keeping styles consistent and making targeted edits. The model can also edit an existing photo without changing the rest of the image. If you know how ChatGPT works, you will recognise the principle: the system predicts the most probable visual result based on your input.
How do you make an image in ChatGPT step by step?
Making an image in ChatGPT starts with opening a new conversation. Log in at chatgpt.com or open the app, and type a description of what you want to see. ChatGPT automatically recognises that you want an image and generates it.
Write a description that is as specific as possible. Name the subject, the style, the colours, the composition and the mood. ‘A modern city at sunset with glass buildings and a pink sky’ gives a more usable result than ‘a nice picture of a city’.
Once generated, the image appears in your conversation. You can download it straight away, or ask ChatGPT for changes. ‘Make the sky darker’ or ‘Add a person in the foreground’ are examples of follow-up prompts that refine the image further. By refining step by step, you get closer to the result you had in mind.
ChatGPT has two places where you can make an image. The first is in every conversation: click the plus button on the left of the chat bar and choose Create image, or just type your description in the text box. ChatGPT then switches on the image feature automatically, and the image appears in your conversation, together with the context you gave there earlier. The second place is Images in the sidebar, a separate image environment that OpenAI introduced on 16 December 2025, on chatgpt.com and in the mobile app.
Experiment with styles in Images; stay in your conversation when the image belongs to an ongoing task.
The difference lies in where you start. In a conversation, you build the image from what you have already discussed, for example a visual for a LinkedIn post you have just written together. In Images, you start without a conversation around it and choose from dozens of preset filters and prompts, which OpenAI regularly updates based on trends. If you want to experiment freely with styles, start there. If the image belongs to an ongoing task, stay in your conversation. In both cases you do not need to choose a separate model.
Image prompt tips for better results in ChatGPT
The quality of your prompt determines the quality of your image. When generating images, ChatGPT makes the same choices as with text: it fills in missing details itself based on probability. The more context you give, the less the model has to guess.
Name five elements in your prompt: the subject (what is central), the style (photorealistic, illustration, cartoon, oil painting), the composition (close-up, bird’s-eye view, symmetrical), the lighting (natural light, studio lighting, dramatic contrast) and the mood (warm, businesslike, playful). Not every prompt needs all five, but the more you name, the more predictable the result.
Avoid vague adjectives such as ‘beautiful’ or ‘professional’. Those terms are subjective and give the model little to go on. Instead, describe exactly what you want to see: ‘a white background with a soft shadow’ is more useful than ‘a professional look’.
Work iteratively. Generate a first version, assess what is right and wrong, and refine your prompt based on the result. This process of framing, output and checking is the same approach that is central to working with text in LearnLLM’s ChatGPT Course. With images, it works no differently.
Write your prompt in English if you notice that results in another language are less accurate. The model was trained mainly on English-language data, so English descriptions are sometimes interpreted more precisely. This applies particularly to specific style descriptions and technical terms.
Editing images in ChatGPT
As well as making images, ChatGPT can edit existing ones. Upload an image with the plus button in the chat bar, under Add photos & files, and give an instruction for the change. The model adjusts only the requested part and leaves the rest of the image intact, including lighting, composition and facial features.
This feature is called ‘inpainting’ and works by selecting a specific area of the image and regenerating it based on your instruction. For example, ask it to change the background, add or remove an object, adjust clothing or put text on a product.
Bear in mind that ChatGPT does not work with your original file, but generates a new version based on your instructions. With product photos or official imagery, this can produce subtle differences. Always check the result carefully before you publish it.
Is ChatGPT image generation free?
Yes, you can make images with ChatGPT for free. The free account gives access to the image feature, but OpenAI describes this on its pricing page as a limited ability to create images. OpenAI does not publish how many images that means per day.
With ChatGPT Go (€8 a month), according to OpenAI, you get more image generation than with a free account. For ChatGPT Plus (€23 a month), OpenAI mentions more complex and more accurate image generation. For Pro, such as ChatGPT Pro 200 (€229 a month), OpenAI promises unlimited and faster image creation, as long as there is no abuse.
The free version is suitable for exploring the feature and judging whether the quality fits your work. In practice, regular professional use requires a Plus subscription.
How many images can you make with ChatGPT?
The number of images you can make with ChatGPT depends on your subscription. OpenAI does not publish fixed numbers per day or per hour. The pricing page only describes roughly how much room you get on each plan, from a limited ability on the free account to unlimited on Pro.
When ChatGPT Images 2.5 arrived, OpenAI said the existing limits for image generation would not change. In cases of abuse, OpenAI can still impose restrictions, even on Pro.
Every generation attempt counts towards your limit, even if the result is not usable. So prepare your prompts well, and avoid resending a prompt while ChatGPT is still generating. That costs an extra attempt without a result. Complex prompts can take up to 2 minutes to process, so wait for the result before trying again.
During peak hours, OpenAI may temporarily lower the limits. There is no dashboard showing how much of your quota you have used, so you only notice the limit when you get the error message.
Limitations of ChatGPT image generation
ChatGPT’s image generation has limitations you need to know about before using it for professional work. The results are not always flawless, and certain types of images are deliberately excluded.
According to the system card for native image generation, OpenAI does not block images of adult public figures by default. Public figures can, however, opt out, and images of minors are blocked. ChatGPT does refuse images in the style of a specific living artist, to protect creators’ intellectual property.
Realistic human faces remain a weak spot. ChatGPT can generate convincing portraits, but errors regularly occur in details such as hands, fingers and teeth. Always check generated images of people carefully. Alternatives are drawn figures, silhouettes or abstract representations.
Text in images has improved considerably since GPT Image 1.5, but it is not yet flawless. Short words and phrases work well, but longer texts or specific fonts can produce spelling mistakes or distortions. Always check text in generated images letter by letter.
Imagery that is entirely generated by a model falls under AI art, a category with its own questions about copyright and provenance.
ChatGPT generates images; it does not take photos. Every image is a new creation based on patterns, not a reproduction of reality. Do not use generated images as a substitute for real product photos in an online shop, and be open with your audience when you use AI-generated images.
What can you use ChatGPT images for at work?
ChatGPT images can be used for a wide range of professional tasks where you need visual material quickly. The image feature is no substitute for a graphic designer, but it can speed up a large part of the visual groundwork.
Social media content is one of the most common uses. Generate a visual for a LinkedIn post, make variations for Instagram Stories or create a banner for a newsletter, without opening a separate design program. Include the format and style you want in your prompt for the best result.
For presentations and reports, you can quickly generate illustrations, concept visualisations or schematic representations. Upload a sketch or describe a concept, and ChatGPT generates a visual you can use as a starting point or as the final image in an internal presentation.
Product visualisations are useful for exploring prototypes, packaging concepts or mood images before you invest in a photo shoot. Use them for internal communication or as a briefing for a designer, not as a replacement for real product photos on your website.
The image feature also adds value for educational content, infographics and blog illustrations. The combination of speed and flexibility lets you create images that fit the message exactly, without depending on stock photos.
Comparison table: image generation by plan
| Feature | Free | Plus | Pro 200 |
|---|---|---|---|
| Generate images | Yes | Yes | Yes |
| Amount according to OpenAI | Limited | No number published | Unlimited, barring abuse |
| Edit images | Yes | Yes | Yes |
| Image model | ChatGPT Images | ChatGPT Images | ChatGPT Images |
| Generation speed | Not stated | Not stated | Faster according to OpenAI |
| Inpainting (targeted editing) | Yes | Yes | Yes |
| Preset styles and filters | Yes | Yes | Yes |
The Pro 200 column refers to the €229-a-month plan. OpenAI also offers Pro 100 (€103 a month, since 9 April 2026), Pro 500 ($500 a month, since 29 September 2026, only known in dollars), ChatGPT Go (€8 a month) and Business (€26 per user per month, €21 when billed annually). Their image limits are not included in this table.
As well as text, LearnLLM’s ChatGPT Course covers working with visual input and output, including workflows for image generation and checkpoints for professional use. Want to get consistently better results from ChatGPT? Sign up for the ChatGPT Course and work with repeatable methods that improve your output.
Frequently asked questions about making images with ChatGPT
If you use ChatGPT image generation for business, you often also want to know what you can and may do with the image afterwards.
Can you tell that an image was made with ChatGPT?
Yes, according to OpenAI, an image made with ChatGPT contains both C2PA metadata and an invisible SynthID watermark. With the verification tool at openai.com/verify you can check whether an image comes from OpenAI. Metadata can be removed, whereas according to OpenAI the watermark is more likely to survive a screenshot, for example. So if the tool finds no signal, that does not prove the image did not come from OpenAI.
Can you use images from ChatGPT commercially?
Yes, you may use images from ChatGPT for business, because according to OpenAI’s terms of use you own the output, to the extent the law allows. For commercial use, OpenAI’s additional business terms also apply. You remain responsible for making sure an image does not break any law or rules, and OpenAI points out that other users may get similar output. So an AI image is not an exclusive image.
Can ChatGPT turn a sketch into an image?
Yes, since ChatGPT Images 2.5 (8 September 2026), you can turn a sketch into an image in ChatGPT with the Sketch feature. In the mobile app, type @ in the message box, choose Sketch, draw a rough layout as a visual guide and then describe the image. ChatGPT uses that drawing as its basis. You then adjust the result with ordinary follow-up prompts.
