How to create images with ChatGPT: from a clear idea to a useful image

Choose the image's job before describing its style

Start with why the image is needed. A banner may need to catch attention and leave room for a heading. A teaching illustration may need to explain how objects fit together. A product concept may need to explore a shape or material. The job controls the details you should ask for.

Try writing one sentence: 'This image should help the reader understand...' or 'This image will appear beside...' If you cannot finish it, the brief may still be too vague. 'Make something beautiful about productivity' leaves many choices open. 'Show a tidy weekly planning desk with a single notebook and room for a title' gives the picture a usable purpose.

Match accuracy to the job. An abstract background can be loose. A diagram explaining a process needs the right steps. A real product page needs a truthful picture of the actual item. A generated concept should not be presented as proof of what a real interface, place, or product looks like.

Start image creation in your available interface

Ask ChatGPT to create an image in a chat where image generation is available. Describe the image rather than asking it only to write a prompt. For example, 'Create an overhead illustration of a lunchbox' makes the intended action clear. You can then continue with editing requests in the same conversation.

Tools and viewing controls can differ between web, desktop, and workspace settings. Current official guidance supports describing an image in natural language and attaching references when they should guide an edit. Follow the controls actually shown in your account rather than expecting every screenshot from another device to match.

If you receive only a written description, clarify that you want an image file generated. If the tool is unavailable, check your account's current options and any displayed message. Repeating the same request with more dramatic wording will not make a missing tool appear. You can still prepare the brief while you resolve access or choose another supported image workflow.

Write a brief with concrete visual details

Name the main subject first, then describe its position and surroundings. 'An open blue lunchbox in the center, viewed from above, on a pale kitchen table' is easier to picture than 'a wonderful healthy lifestyle image.' Add the visual details that determine success, such as a clear lid, separate food compartments, or space around the edges.

Describe the layout in everyday words. Say what should be in the foreground, at the back, on the left, or in the center. If a heading will be placed over the image, specify the empty area. If the image will be wide, say where the subject should sit so a mobile crop will not cut it away.

Use a small number of style choices that work together. A soft paper illustration, a realistic studio photo, and a flat diagram are different directions. Combining every style word you like can produce a muddled result. Choose the direction that helps the reader and name colors, light, or materials only when they matter.

Use reference images with a clear role

Attach a reference when you need ChatGPT to edit a picture or follow an existing visual direction. Explain what the reference is for. It might define the subject, color palette, shape, or arrangement. Without that explanation, the model has to guess which features you want to preserve.

If you provide several references, label their jobs. For example: 'Image 1 shows the object and camera angle. Image 2 shows the color palette. Keep the object from image 1 and apply only the palette from image 2.' This is clearer than asking to blend everything together.

Use images you are allowed to supply and use. A reference does not automatically grant permission to publish a person's likeness, a protected design, or a brand asset. For factual material, keep the real source available. When accuracy is essential, editing a real authorized image may serve the task better than generating an imagined replacement.

A complete lunch-guide image example

This is a teaching example of a brief and review process, not a claim that a particular image was generated or tested. Suppose you need a wide image for a lunch-planning article. The page heading will sit in the top-left, and the article will explain the food choices in ordinary text.

A suitable request is: 'Create a wide editorial illustration for a simple lunch-planning guide. Show one open lunchbox from above with a sandwich, a small fruit compartment, and a water bottle beside it. Use a pale cream background and muted blue and green colors. Put the lunchbox slightly right of center. Leave the top-left area clear. No people, logos, or readable text.'

Imagine the first result has two lunchboxes and the bottle covers the fruit. The useful correction is: 'Keep the illustration style, colors, and overhead view. Remove the second lunchbox. Move the water bottle to the right so every food compartment is visible. Keep the top-left area empty.'

The acceptance check is concrete: one lunchbox, visible compartments, an uncovered fruit portion, and clear space for the heading. If the image passes those checks, further edits should serve a real need. You do not have to chase a different style simply because another variation is possible.

Edit one main problem while preserving the good parts

Identify the largest mismatch before asking for a new version. It might be the wrong subject, an awkward crop, an extra object, or text that is hard to read. Fix the thing preventing the image from doing its job. Fine color changes can wait until the content and composition are right.

Say both what changes and what stays. 'Make the bottle smaller; preserve the lunchbox, colors, angle, and empty area' gives the edit a boundary. 'Improve this' could lead to a whole new composition. Even with a clear request, inspect the full image afterward because an edit can affect nearby details.

Where your interface supports selecting or commenting on an area, use that to point to the issue. Keep the written instruction precise. For a large style change, state that the overall look may change but list the objects and layout that must remain. Save a useful version before major edits so you can return to it if needed.

Handle words, diagrams, and factual details carefully

If the image needs words, keep the text short and provide the exact spelling, capitalization, and placement. Check every letter in the result. An image can look convincing while containing an incorrect word, a repeated label, or tiny text that cannot be read at its actual size.

For a website or document, consider placing important words outside the image. Normal page text is easier to update, select, translate, and read with assistive tools. A heading added by the page layout can stay sharp across screen sizes. The illustration can carry the visual idea while the page carries precise instructions.

Use ordinary diagram tools or code when a chart must show exact numbers, connectors, or labels. Image generation can help explore the look, but a factual diagram needs a method that keeps relationships correct. If a generated picture includes a measurement, scientific structure, interface, or map, verify it against a reliable source before presenting it as accurate.

Check the image at the size and crop you will use

Open the final image large enough to inspect details, then view it at its real display size. A flaw visible only at extreme zoom may matter less than a subject that disappears on a phone. Check the actual page or document when possible, not just the full-size image viewer.

Cropping removes part of the image to fit a frame. A wide image may be cropped more tightly on a small screen. If the lunchbox is pushed too far right, a centered crop might cut it off. Ask for a more central arrangement or change how the page places the image.

Check the contrast and nearby text too. Busy detail behind a heading makes the words harder to read. Empty space is useful only if it is where the page needs it. For a printable piece, check the output size and detail before ordering a large run. A draft preview is the time to catch these problems.

Choose a practical file and describe it accessibly

Download the final version through the interface and confirm that the file opens. Check its dimensions and format in the destination application. An image suitable for a small web card may be too small for a large print. An image with a background may not work where you need a transparent cutout.

If you need transparency, request it clearly and verify the downloaded file in a viewer that shows transparent areas. A checkerboard pattern can be part of the picture rather than actual transparency. Look at the file itself, not only the preview.

Give the image useful alternative text when the destination supports it. Describe the information it adds: 'Open lunchbox with separate sandwich and fruit compartments, with a water bottle beside it.' Avoid a long account of decorative colors unless they matter. If the image is only decoration, follow the destination's accessibility approach for decorative images.

Know when a different method will give a better result

Choose generation for illustration, mood, ideas, and visual exploration. Use a real authorized photo when the reader needs to see an actual product or place. Use a verified screenshot for a real interface. Use charts or diagrams built from the source data when exact values or relationships are the point.

For recurring images, keep a short style brief and a few approved references. Reuse the palette, level of detail, and spacing so the collection feels consistent. Change the subject and composition to fit each piece rather than making every image a copy of one cover.

Keep a record of the final brief and important edits when you expect to reuse the image. The goal is a useful visual you can place confidently, not an endless set of variations. Finish when the image supports the content, survives the intended crop, and meets the accuracy and file checks for its real destination.

After choosing the image, write a description suited to where it will appear. Describe an image clearly →