ChatGPT imagesWorkflow & organisationConsistent characters
ChatGPT Image Editor: Select, Edit and Change Aspect Ratio
To edit an image in ChatGPT, select it to open the editor, click Select and highlight the area to change, then describe the change and send. You can also skip the selection and describe the edit in the conversation, naming the area in your prompt. The editor's Aspect ratio option regenerates the image in a new shape; on mobile you also get Resize and Remove. Highlights are not always exact, so keep edits small.
Editing used to mean asking ChatGPT to redraw the whole picture and hoping the rest survived. The editor in ChatGPT Images changes that: you can mark the part to change and leave the rest alone, at least most of the time. It works on images ChatGPT made and on images you upload. This guide covers the editor on web and mobile, how to phrase edits so they stay local, and what to do when an edit spreads further than you wanted.
Opening the editor
Select an image in your conversation, or open it from the Images library. You can edit images generated by ChatGPT Images or upload an existing image and describe the changes. OpenAI describes two ways to edit:
- With Select: highlight the part of the image, then describe the change in the editor text box or the chat.
- Without Select: describe the edit directly in the conversation. If the edit should apply to a specific area, name that area in the prompt.
The editor's tools
| Setting | What it does | Start with |
|---|---|---|
| Select | Highlights the area to edit | Cover the object plus a small margin |
| Aspect ratio | Regenerates the image in a different shape | Use before detailed edits |
| Undo / Redo | Revises your selections | Tidy the highlight before sending |
| Cancel | Starts over | When the selection went wrong |
| Save | Keeps the edited version | After checking at full size |
| Resize, Remove (mobile) | Extra image controls in the app | Quick fixes on the go |
OpenAI's note is worth taking seriously: highlights are not always precise, and edits may extend beyond the area you selected. The developer documentation says the same about masks for the underlying models: masking is prompt-based, and the model uses the mask as guidance rather than following its exact shape.
Editing on mobile
In the ChatGPT app, tap an image generated by ChatGPT Images to open it full screen. Tap Edit to describe changes in the conversation panel, or Select to highlight areas: a slider changes the size of the selection brush, with Undo and Redo. Tap Next, then describe the changes for the highlighted areas. You can also add comments to an image on mobile.
Writing edits that stay local
- Name the object and the change: "Change the mug in the selected area to matte black ceramic."
- Say what must not change: "Keep the hand, the table and the lighting exactly the same."
- One change per edit. Two edits in sequence are easier to control than one combined request.
- Match the existing light. "Same soft window light from the left" stops a new object looking pasted in.
- Edit from the best version. If an edit drifts, go back to the previous approved image rather than editing the edit.
Changing the aspect ratio
ChatGPT Images can generate any aspect ratio, chosen with the picker or stated in the prompt. In the editor, the Aspect ratio option regenerates an existing image in a different shape, useful when a square social image also needs a 9:16 story version. Because it regenerates, details can change; do it before fine edits, then make your detailed corrections on the new version. For the right shape per platform and size limits, see ChatGPT image sizes and quality.
Keeping a character or product consistent
Edits are also the most reliable way to keep a subject consistent: change the scene around an approved character or product rather than generating it again. Upload the reference, describe the new setting, and list the features that must stay identical. OpenAI's documentation lists consistency of recurring characters and brand elements across generations as a remaining limitation, so expect to choose between versions and to make small fixes. More on this in edits and references.
A worked editing sequence
Suppose a client approves a lifestyle image of a coffee table but wants the mug changed to their product, the magazine removed and a warmer evening feel. Doing all three in one request invites the model to redraw the whole scene. Instead:
- Select the mug and replace it with the product, using the product photo as a reference and listing what must stay identical.
- Check the result at full size; if the label is off, correct only the label.
- Select the magazine and ask for "the bare wooden table surface, matching the grain and light".
- Finally, without a selection, ask for "a warmer, late-afternoon light across the whole image, keeping every object the same".
Global changes such as light or colour belong last, because they touch every pixel. Save each approved step so you can return to it if a later edit goes wrong.
Fixing common problems
- The edit spilled outside the selection. Make the selection tighter, add "change only the selected area" and list what to keep.
- Text changed when you edited something else. Name the text and say it must stay exactly as it is.
- The whole image shifted colour. Add "keep the colour grading and lighting identical".
- Nothing changed. Describe a visible result, not a vague wish: "replace the plant with a tall cactus" rather than "make it nicer".
Try it
In the selected area, replace the potted plant with a small stack of three hardback books in muted green and cream. Keep the desk, the lamp, the window light and everything else exactly the same.
Generate or upload a desk scene first, select only the plant, and send this edit. Then try the same request without Select and compare how much of the rest of the image changes.