Image-to-Image Prompt Guide: How to Write Edit Prompts That Work

A good edit prompt says what to change first, then what must stay the same. One or two plain sentences usually beat a long paragraph. If you upload more than one image, refer to them as "Image 1", "Image 2" and give each one a job.

This guide covers the pattern, the recipes we see most often, and what to do when the result drifts.

The basic pattern

[The change]. [Where or how, if it matters]. Keep [what must not change].

Example:

Replace the background with a quiet beach at sunset. Keep the person,
her pose, clothes and face exactly the same.

1. Lead with the change

Put the verb first: replace, remove, add, change, turn, restyle, colorize. The model reads the whole prompt, but an instruction buried after three sentences of description is the one most often ignored.

Be specific about the result, not the effort. "Make it better" gives the model nothing to do. "Brighten the room, warm white light from the window on the left" does.

Google's Gemini API prompting guide gives the same advice for Nano Banana: be specific, and say what the image is for. "A product photo for an online shop" steers the result more than a list of adjectives.

2. Name what to keep

An edit model doesn't know which parts of your photo matter to you. Say it:

• People: face, hairstyle, expression, pose, clothes
• Products: shape, colour, logo, label text
• Scenes: composition, camera angle, lighting

"Keep everything else the same" is a useful last line. "Change only the shirt" is even clearer when the edit is small.

3. One change per pass

Several changes at once ("new background, new outfit, make it night, add a dog") multiply the chance that something drifts. Do them in steps: download the result, upload it again and make the next change.

Multiple images: Image 1, Image 2

Uploads are numbered in the order you add them: the first is Image 1, the second Image 2, and so on. Type @ in the prompt box to insert a name. Nano Banana 2, Nano Banana Pro and GPT Image 2 take up to 14 images here; Seedream 5.0 Flash takes up to 10.

Give every image a job. Without one, the model guesses what each upload is for.

Put the woman from Image 1 in the jacket from Image 2. Keep her face,
hair and pose from Image 1. Keep the jacket's colour, cut and buttons
from Image 2. Plain light-grey studio background.

Put the photo you're editing first. With the aspect ratio on auto, the result takes the shape of Image 1.

Aspect ratio: leave it on "auto"

In edit mode, auto keeps the shape of Image 1, which is what you want for most edits. Pick a ratio only when the format must change, for example a square product shot turned into a 9:16 story. The model then has to invent content for the new space, so say what goes there:

Extend this photo to a vertical 9:16 frame. Continue the sky above and
the wooden deck below. Keep the person and the table unchanged.

Recipes

Background swap

Replace the background with a bright, minimal studio backdrop in light
beige, with a soft shadow under the subject. Keep the subject exactly
the same.

Describe the light so the subject fits in. See the background changer guide for more.

Restyle

Turn this photo into a watercolor illustration with loose brush strokes
and soft paper texture. Keep the composition and the people's poses.

Name the style with what you'd see (brush strokes, paper, flat colours, film grain) rather than only a label.

Outfit change

Change his shirt to a navy linen shirt with the top button open. Keep
his face, hair, pose and the background unchanged.

For a specific garment, upload it as Image 2 and use the two-image prompt above.

Product scene

Place the perfume bottle from Image 1 on a wet black stone surface with
a few drops of water around it, dark green leaves in the background,
soft light from above. Keep the bottle's shape, cap and label text
exactly as they are.

Several products in one shot

Arrange the mug from Image 1, the notebook from Image 2 and the pen
from Image 3 on a light oak desk, seen from above. Morning window light.
Keep each product's colours and logos.

Remove objects

Remove the people in the background and the trash bin on the right.
Fill the space with the same street and buildings. Keep the couple in
the foreground unchanged.

Say what should fill the gap. "Remove" alone sometimes leaves a blur.

Colorize and restore an old photo

Colorize this black-and-white photo with natural skin tones and muted
1960s colours. Remove the scratches and dust. Keep the faces and
expressions exactly as they are.

Change the light or time of day

Make it night: street lamps on, warm light from the shop windows, dark
blue sky. Keep the buildings, the car and the camera angle the same.

Replace text in an image

Change the text on the shop sign to "BLUE DOOR CAFE" in the same font
style and colour. Keep everything else the same.

Put the exact words in quotes and keep them short.

Troubleshooting

The face doesn't look like the person anymore (identity drift)

• Say it outright: "keep her face exactly as in Image 1".
• Make one change per pass instead of several.
• Use a photo where the face is large, sharp and evenly lit. A face that is a few pixels wide in a group shot has little detail to keep.
• Render at 2K. More pixels on the face means more detail kept.
• For several people in one image, try Nano Banana Pro. Google says it keeps the consistency and resemblance of up to five people (Google).

Spelling mistakes in text

• Put the exact text in quotes: the label reads "MATCHA LATTE".
• Keep it short. A headline renders more reliably than a paragraph.
• Say where it goes and how it looks: "centred at the top, bold white sans-serif".
• Use Nano Banana Pro or GPT Image 2 for text-heavy designs.
• Fix one word at a time: "Change only the word 'Moring' to 'Morning'."

The change didn't happen

Move the instruction to the start, make it concrete, and remove anything that contradicts it. "Keep everything the same" plus "change the background" works; "keep the photo exactly as it is" plus a big change doesn't.

Too much changed

Add a keep list, or narrow the edit: "change only the sky".

The wrong image was used

Name each image's job ("the background from Image 2, the person from Image 1") and check the upload order.

Which model?

Nano Banana 2 is the default and handles most edits. Nano Banana Pro is slower and costs more, and handles text and multi-person scenes best. GPT Image 2 is good at precise edits and clean typography. Seedream 5.0 Flash is the cheapest way to test an idea. Prices and details are in our model comparison.

Checklist

1. Start with the change, as a verb.
2. Say what must stay the same.
3. One change per pass.
4. Name every upload: Image 1, Image 2.
5. Aspect ratio on auto unless the format must change.
6. Text in quotes, short.

New accounts get 10 free credits, which is five Nano Banana 2 images at 1K, and failed images are refunded. Open the editor and try one of the prompts above.