The ChatGPT image generator can create a new picture or revise an image you upload. OpenAI’s current help names the feature ChatGPT Images; you can request an image in a conversation or open More, then Images. Start with one clear visual purpose and keep an accepted version before asking for changes.
Give the image one job
Imagine a cover illustration for a fictional mountain café: an amber mug beside a window, a pine forest outside, and empty space at the top for a heading. The image’s job is atmosphere and composition. It does not need to invent a finished logo, a price list and a realistic map at the same time.
An original starting brief would be: “Create a square illustration of an amber ceramic mug on a wooden café table. A pine forest is visible through the window. Keep the upper quarter visually quiet for a heading I will add later. Use a restrained green and amber palette.”
Choose a fictional or properly licensed subject for a first attempt. Avoid putting private documents, client material or another person’s likeness into an experiment without checking whether you have permission and whether the service’s data settings meet your needs.
Use a change-versus-preserve worksheet
| Decision | Café example | What to inspect |
|---|---|---|
| Subject | One amber mug | Shape, handle and count remain plausible |
| Composition | Quiet area above the table | The crop leaves space for the planned heading |
| Visual treatment | Restrained green and amber illustration | The result suits the intended page |
| Revision | Change the outdoor light to dusk | The mug and layout do not change unintentionally |
This worksheet makes the second version easier to judge. “Make it better” gives no boundary for improvement. “Make the outside scene dusk while retaining the mug, table and framing” identifies the intended difference and the details you will compare afterward.
Separate aesthetic preference from a requirement. A different shade of green may be acceptable. A second mug, an unreadable heading or an incorrect product detail can make the asset unusable even when the overall image is attractive.
Create and revise in ChatGPT Images
OpenAI’s Images guide documents conversational creation, uploaded-image edits and a selection tool for local changes. It warns that a selected edit can extend beyond the highlighted area. An aspect-ratio control can regenerate the image in another shape; that is a new version to review, not merely proof of a lossless crop.
- Write the brief and choose the required shape. Review the first result against the worksheet before requesting decorative changes.
- Save an acceptable baseline. Ask for one consequential revision at a time, stating both the desired change and the elements that should remain.
- For a local correction, select the relevant area or describe it explicitly. Inspect nearby details as well as the edited region.
- Compare the result with the baseline. Keep the version that meets the brief; more iterations do not automatically mean a better asset.
Three common revisions require different judgments. Changing the atmosphere affects the whole picture. Removing an object needs attention to the newly filled space. Correcting words requires checking the exact characters. Treating all three as a general request to “improve” the image makes unintended changes harder to spot.
| Requested revision | Accept when | Return to the previous version when |
|---|---|---|
| Change daylight to dusk | The mood changes and the important shapes still read | The mug or furniture changes without a reason |
| Remove a spare spoon | The table surface looks coherent | The remaining object boundaries become distorted |
| Leave more title space | The required blank region exists at the final crop | The main subject becomes too small to communicate |
Uploading an image is different from generating one
The image-input FAQ describes adding an existing image through the attachment control, drag-and-drop or paste. Those inputs allow visual discussion as well as editing; an analysis response is not a newly generated asset. State whether you want a description, a proposed change or an actual image revision.
For the café example, an uploaded sketch can establish where the mug and window belong. It need not be polished. Its value is making the layout requirement explicit. If a precise commercial object must remain unchanged, inspect the generated revision closely; a visual reference does not establish exact reproduction.
Understand limits before planning a batch
OpenAI’s Free Tier FAQ confirms that image creation has usage limits separate from ordinary text chat. A successful text conversation therefore does not establish remaining image capacity. Use the account’s notice for the applicable reset or upgrade options rather than assuming an old daily image count still applies.
For a small project, decide how many meaningful revision questions you need to answer. A sensible first session might establish composition, test one edit and save an accepted version. If the allowance ends before that, record what you learned; do not interpret a limit as evidence that the prompt was wrong.
Save the version you intend to use
Generated images are available in the Images area. OpenAI documents Save for downloading, as well as Copy and Share. Deleting the originating conversation removes its images from My images. Keep the file you need outside the conversation before relying on it as your only working copy. Image management instructions
Open the download at its intended display size. Check the actual dimensions, spelling, subject details and crop. Add final layout text in an editor when exact typography matters. The generated visual and the finished publication asset are separate deliverables.
When you want a character you can interact with
A portrait is one asset. A companion character also needs personality, a scenario and a conversation opening. Cherrypop’s character creator combines those choices with appearance, with realistic and anime options. That is a more focused starting point when interaction is the goal, rather than a café illustration or marketing layout.
You can then explore the supported image and video creation flows around the character. This is a different workflow, not a claim that Cherrypop replaces ChatGPT’s general image editor or guarantees identical appearance across every output.
Cherrypop is free to start with limits; some features require Premium or Cherries. If the image is the beginning of a companion idea, build the character and its scenario before choosing the next piece of media.