Gemini can edit an image you upload, revise an image it generated, or combine several uploaded images into a new scene. The result depends heavily on the editing brief: a useful Gemini image editing prompt says what must change, what must stay, and how the finished frame should look. Below are eight ready briefs by type of edit.

For Google’s own edit templates, community prompts and fixes for bad edits, see the Gemini prompt for photo editing guide.

Docs checked: 26 October 2026. Google says image editing in Gemini Apps requires signing in and is available to users aged 18 or over in supported countries and languages. Product labels, access and limits can change.

On this page

Start with a controlled edit

Upload a photo you have permission to use, open it in Gemini and choose Chat. Google’s help page says you can then describe the change you want. For a new image, Google recommends beginning with a direct verb such as “draw”, “generate” or “create”, naming the visual style and describing the subject, action and background. Where the image tools sit and which model each plan gets is covered in how to use Nano Banana.

Editing needs one more instruction: name the parts that should remain unchanged. Without that boundary, a request to replace the background can also alter clothing, lighting or facial details.

Use this base structure:

Edit the uploaded photo. Change [one specific element] to [desired result]. Keep [subject, pose, clothing, crop and other fixed details] unchanged. Match the existing [light, perspective, shadows and colour]. Return [aspect ratio or output use].

Treat the output as a new draft, not a corrected original. Check names, labels, hands, reflections and small objects before you publish it.

1. Replace a distracting background

Edit the uploaded photo. Replace the current background with a quiet home office: warm off-white wall, oak desk and one small green plant. Keep the person, pose, clothing, hair and facial features unchanged. Match the existing direction of light and preserve the natural shadow around the subject. Keep the same landscape crop.

This prompt gives Gemini a small scene rather than asking for a vague “professional background”. If the subject changes, follow with: “Restore the subject from the uploaded photo exactly; change only the background.”

2. Remove one unwanted object

Edit the uploaded photo. Remove the red bag beside the chair. Reconstruct the floor, skirting board and chair leg behind it so their lines and texture continue naturally. Do not change the person, furniture, crop, colour balance or lighting.

Name the object by colour and position. “Remove the clutter” leaves the model to decide what counts as clutter, and it may remove useful objects too.

3. Change the colour of a product

Edit the uploaded product photo. Change only the blue fabric panel to deep forest green. Preserve the seams, weave, printed logo, highlights, shadows, hardware and background. Keep the product shape and camera angle exactly the same.

Colour changes are easier to review when the prompt lists the material details that carry shape. Compare the original and edited edges at full size. A clean thumbnail can hide warped stitching or a changed logo. Google’s own version of this edit is in the library: change only the sofa in a living room photo.

4. Reframe for a social post

Extend the uploaded image into a vertical 4:5 frame. Keep the original subject at the same scale and place it in the lower centre. Continue the wall and floor naturally above and at both sides. Leave clear, uncluttered space in the upper third, but do not add text, logos or new objects.

This is an outpainting brief: it tells Gemini where to add material without asking it to redesign the source. If you need both portrait and landscape versions, request them separately and check each composition.

For more control over the frame, see the Midjourney aspect ratio guide. The principles of ratio and composition still apply even though the controls differ.

5. Adjust light without changing the scene

Edit the uploaded interior photo to look as though it was photographed in soft morning daylight from the window on the left. Lift the dark shadows slightly and reduce the yellow cast. Keep every object, surface, material, reflection and camera position unchanged. Do not add sunlight beams or move furniture.

“Make it brighter” can flatten the scene. Direction, softness and colour give the model a clearer target. Do not describe a lighting setup that conflicts with visible windows or shadows.

6. Combine two reference images

Upload both images, then write:

Create one new image using the person and clothing from image 1 in the room from image 2. Place the person beside the window, standing naturally and looking at the desk. Match the room’s perspective, colour temperature and shadow direction. Do not copy text, signs or watermarks from either image. Use a landscape 16:9 frame.

Google says Gemini can use multiple uploaded images to create a new image. Identify each reference by number and assign it one job. This is more precise than “combine these pictures”, which does not say which composition or details should win.

Only use people and source images you have permission to edit. Do not use a prompt to imply a real person attended an event, endorsed a product or appeared somewhere they did not.

7. Turn a photo into a specific illustration

Create an editorial illustration based on the uploaded photo. Use thick black contour lines, flat warm cream and burnt-orange shapes, sparse detail and no gradients. Preserve the subject’s pose, the position of the main objects and the wide landscape composition. Simplify the background. Do not add text.

Google’s general prompting tips recommend naming the style. Go beyond “cartoon”: describe line weight, palette, level of detail and what should carry over from the photo. If you need a more literal likeness, say which features matter and avoid asking the model to exaggerate them.

8. Make one precise follow-up edit

Use the most recent image. Change only the mug from orange to white. Keep its shape, handle, position, reflection and shadow unchanged. Do not alter any other part of the image.

Follow-up prompts work best when each one has a single testable change. If a draft needs five corrections, make the most important correction first. Recheck the whole image after every revision because an apparently local edit can affect nearby details.

A short review prompt

After an edit, you can ask Gemini to help you inspect the draft, but you still need to look at the image yourself:

Compare the edited image with the uploaded original. List any unintended changes to the subject, text, logos, hands, reflections, object count, perspective, crop or lighting. Do not claim a detail is correct if it is unclear.

This is a review request, not proof. Zoom in on any area that matters to identity, product accuracy or publication.

What to include in every photo editing prompt

A reliable brief answers five questions:

  1. What source image should Gemini use?
  2. What single change is required?
  3. Which details must remain fixed?
  4. What lighting, perspective and material cues must match?
  5. What crop or destination will the image serve?

The wording does not need to be long. It needs to remove the decisions you do not want the model to make.

Sources