ChatGPT image generator: create, refine, and keep versions

To use the ChatGPT image generator, describe the image in a chat or choose More > Images, then refine the result with editing instructions. OpenAI documents creation, editing, downloads, and saved images in Images in ChatGPT↗ (opens in a new tab). fenbs can keep your prompts, version notes, and approved direction so you can continue the work in another AI later.

A useful workflow has three parts: a clear first request, a focused change, and a record of what you kept. Start with one image for one purpose. You can explore alternatives after you have a result that meets the basic requirements.

Where do you start creating an image?

These instructions cover the consumer controls on web, iOS, and Android documented in the Images help article linked above. Follow the instructions for your own product surface when using another client; do not assume every surface has the same controls.

1. Open a ChatGPT conversation and ask it to create an image, or choose More > Images.
2. Describe the subject, intended use, composition, and the details the result must preserve.
3. Send the request, then inspect the generated image against your requirements.
4. For a starting design, open Images in the sidebar, choose Templates, and customize a starting design with the details you need.

Templates are not yet available in Work mode, according to the consumer help page↗ (opens in a new tab). Keep that distinction in mind when following instructions written for a different part of ChatGPT. You can still make your request specific without depending on a template.

What should the first prompt include?

Tell the assistant what the image is for before listing visual details. A workshop poster and a website background can show the same subject but need different composition. Write the nonnegotiable requirements first, then describe the style in concrete language.

• Purpose: where the image will appear and who needs to understand it.
• Subject: the main person, object, or scene and what is happening.
• Composition: position, framing, empty space, and the desired aspect ratio.
• Appearance: colors, materials, lighting, and a clear visual treatment.
• Constraints: exact text, excluded elements, and anything that must remain unchanged.

Use a short, original request like the example below. It is a starting prompt, not a tested promise about the generated result. Replace its details with your actual requirements, and inspect what comes back rather than assuming each instruction was followed.

Create a wide illustration for a community repair workshop. Show two adults repairing a wooden chair at a workbench. Use simple shapes, muted blue and warm cream, and soft side lighting. Keep the upper-right quarter clear for text added later. Include no words, logos, or extra people.

When the result is close, preserve the successful instruction. In fenbs, you can keep the prompt and the reason it worked, so the next session starts with a useful direction rather than another vague request.

How do you change one part without losing the direction?

Select a generated image to open its editor. OpenAI documents the Select and Aspect ratio controls and editing through the conversation. Selection-based edits can affect areas outside the highlight, so inspect the entire result after each edit. See the editor instructions↗ (opens in a new tab).

1. Choose the image you want to revise. Identify the exact version in your own notes.
2. Use Select for a local change, or describe the edit directly in the conversation.
3. State one change and list the important details that should remain the same.
4. Compare the new result with the earlier image. Check the background, crop, text, and other areas beyond the intended change.

Change the chair cushion from red to muted blue. Keep the people, chair shape, workbench, crop, lighting, and empty space in the upper-right corner the same. Add no text or logos.

If the revision loses something important, name that difference in the next instruction. Do not call it simply “wrong.” A precise observation, such as a changed crop or an added object, gives you a clearer next test and a more useful note about what happened.

How do you save images and share a useful prompt?

Open Images in the sidebar to revisit generated images. OpenAI says they are saved automatically there, and documents Copy, Save for downloading, and Share on web and mobile. On mobile, Share > Prompt template > Copy link shares a generated image’s prompt so another person can make a version. These controls are described in Images in ChatGPT↗ (opens in a new tab).

Keep downloaded results in your usual asset folder. Give each file a simple name such as repair-poster-v01 and repair-poster-v02, and put the matching instruction in your work notes. That naming scheme is your own organizational method, not an automatic ChatGPT version-control feature.

The same help article says removing an image from My images requires deleting the conversation that created it. Saved images and prompt links do not establish exact reproduction or an immutable version history. Download the versions you need before making cleanup decisions.

What should you check when the result is not usable?

• Wrong composition: describe the position and empty space you need rather than adding more style adjectives.
• Unwanted changes after an edit: inspect the whole image and name what should be restored.
• Text errors: check every visible word against the requested wording before approving the asset.
• Wrong account: confirm you are in the account intended for the work before looking for earlier images.
• Different controls: confirm whether you are following consumer web/mobile instructions or a separate desktop workflow.

For account questions, follow how to switch ChatGPT accounts. For interpreting an image you already have, use can ChatGPT analyze images?. That is a different job from creating and revising a new asset.

For linked teen accounts, OpenAI’s parental-controls guidance↗ (opens in a new tab) includes an Image generation control. If a family setting affects access, see how to set ChatGPT parental controls for that separate account-management question.

ChatGPT’s plugin connection guide↗ (opens in a new tab) documents outside tool connections; account and workspace policies apply. Check the approved connection before using an assistant to help keep your work elsewhere.

When someone approves a version, save what they approved and any remaining conditions. A comment like “looks good” is ambiguous if several images are in play. Attach the approval to your version name and describe its intended use.

How to keep image prompts and versions with fenbs

Never lose the useful work between image sessions. Keep the prompt, the version name, and the next change so you can carry on from exactly where you left off, on any device and in another AI.

1. Create a free account.
2. Add a task for the image you need. Write where it will be used and the details it must include.
3. Connect ChatGPT to fenbs. Use your approved connection and check any workspace restrictions first.
4. Save the prompt, the name of the image file you kept, and what a person approved. Keep the actual image in your usual asset storage.
5. Before you stop, write the next change and any checks still open. Ask your next connected assistant to read the saved work first.

fenbs does not automatically import your image chats or provide an image-editing workspace. Use it for the prompts, decisions, and progress you choose to keep.

If you only need one image and no later handoff, saving the file may be enough. For broader work, browse the AI at work hub. If the images support launch work, ChatGPT for startups covers the surrounding workflow.