Gemini Prompts for Image Editing With Nano Banana

The short version: Nano Banana inside Gemini edits images well when you give it one instruction at a time, name the exact element you want changed, and tell it what to leave alone. Vague prompts like "make this better" get you vague, often ugly results. Specific, boring, almost pedantic prompts get you results you can use.

I've spent a fair bit of time poking at Nano Banana since it landed in Gemini, and the thing nobody tells you upfront is that it's not a mind reader. It's a very capable editor with a very literal brain. Treat it like a new freelancer on their first week who needs clear briefs, not like a design director who already knows your vibe.

Why most Nano Banana edits come out wrong

The number one mistake I see people make with image editing prompts is bundling five changes into one sentence. "Change the background to a beach, make her smile bigger, add sunglasses, fix the lighting, and make it look more professional" is five separate jobs crammed into one prompt. Nano Banana will do a bit of each and nail none of them.

The fix is unglamorous: one edit per prompt, then review, then next edit. It takes longer per image but the success rate per edit goes up massively, probably from something like 40% usable on the first go to closer to 85% when you isolate instructions. I haven't run a formal study, that's a feel from actual use, but the pattern holds every time.

If you haven't used the tool at all yet, it's worth reading how to use Nano Banana in Gemini first, because half of the "bad results" complaints I see online are really just people not knowing the basic create-edit-redo workflow exists.

The prompt structure that works

Here's the format I use, broken into four parts every time:

• Subject - name exactly what's in the image you're editing, not "this picture"
• Action - one verb: remove, replace, add, recolour, extend, blur
• Target - the specific object or area, not the whole image
• Constraint - what must stay the same

So instead of "edit this photo to look more professional," you write: "Remove the coffee cup on the desk in the foreground. Keep the laptop, the lighting and the background exactly as they are." That's a complete, answerable instruction.

A worked example: fixing a product shot

Say you run a small candle business and you've got one decent photo of a candle on a wooden table, but the background is cluttered with a phone charger and a mug. Here's how you'd work through it in Gemini with Nano Banana, step by step:

• Step 1: "Remove the phone charger cable and the mug from the table surface. Keep the candle, the wood grain and the shadow under the candle unchanged."
• Step 2: Once that's clean, new prompt: "Change the table surface from wood to a plain light grey backdrop. Keep the candle and its shadow exactly as they are."
• Step 3: "Add soft, warm studio lighting coming from the top left. Do not change the colour of the candle wax."
• Step 4: "Crop the image to a 1:1 square for Instagram, keeping the candle centred."

Four small prompts, four checkpoints, and by step 4 you've got a usable product shot you can check at every stage instead of hoping one giant prompt magically works. If one step goes wrong, you only redo that step, not the whole chain.

Prompts for specific editing jobs

These are the edit types people ask about most, with a prompt template for each.

Removing objects or people

"Remove [specific object] from [specific location in the image]. Fill the space naturally to match the surrounding [background/wall/floor]. Do not alter anything else in the frame."

Changing backgrounds

"Replace the background with [description]. Keep the subject's pose, lighting direction and edges exactly as they are, so it doesn't look pasted on."

That last clause matters. Without it, Nano Banana sometimes gets the background right but leaves a harsh, cut-out edge around the subject that looks exactly like what it is: a badge sticker job.

Changing outfits or colours

"Change the colour of [specific item] to [colour]. Keep the fabric texture, folds and shadows consistent with the original."

Fixing lighting

"Adjust the lighting to look like it was shot near a window in late afternoon. Keep the subject's position and facial features unchanged."

Combining two images

Nano Banana can take elements from one image and place them into another, which is one of its strongest features, better than most people give it credit for. The prompt needs to specify which image contributes what: "Take the person from image one and place them into the setting from image two, standing near the window on the left. Match the lighting of image two."

The uncomfortable bit nobody mentions

Most posts on this topic will tell you Nano Banana is brilliant at character and object consistency across edits. It mostly is. But it still drifts, especially with faces, after three or four rounds of edits on the same image. You ask it to change a background, then a shirt colour, then the lighting, and by edit four the face has subtly morphed, the nose is a touch different, the eyes sit slightly wrong. Nobody wants to say this because it sounds like criticising a tool everyone's excited about, but if you're doing client work or anything brand-facing, you need to check the face against the original every single edit, not just at the end. Catch the drift early or you'll be starting the whole chain again.

The practical fix: keep your original image saved separately and never edit the edited version more than three times before going back to the source and re-applying your final set of changes as one fresh pass. It's extra admin but it saves you from shipping a slightly-off face to a client who will absolutely notice.

Prompts for batches and consistency

If you're producing a set of images that need to look like they belong together, for example a run of 10 product photos for an online shop, don't prompt each one from scratch. Build your edit chain on image one until you're happy, write down the exact sequence of prompts you used, then apply the same sequence to images two through ten. This is slower than mass-generating but it's the only way to get a consistent set rather than ten images that all look like they came from different photographers.

For social content specifically, pairing your edited images with real caption and content prompts matters just as much as the image itself. There's a full breakdown of that in Gemini prompts for social media content if you want the content side sorted alongside the visuals.

Nano Banana versus Nano Banana Pro for editing

If your editing involves fine text inside images, logos, or multi-step compositing, the Pro model handles it better but comes with tighter usage quotas depending on your Gemini plan. I'd recommend reading which Gemini image model you get and the quota catch before you build a workflow around Pro-only features and then hit a wall mid-project. And if you're weighing it against other tools entirely, the comparison in Nano Banana vs Midjourney is worth a read before you commit your whole workflow to one tool.

Quick checklist before you prompt

• Is this one instruction, or secretly five?
• Have I named the exact object or area, not "the image"?
• Have I told it what to keep unchanged?
• Am I checking the result against the original, not just the last edit?
• If this is a batch, have I written down my prompt sequence to reuse?

Image editing is only half the picture now that Gemini's tools talk to each other. If you're also touching video, the workflow in how to use Gemini Omni to make and edit AI videos by chatting follows the same one-instruction-at-a-time logic, and it's worth knowing before you try to animate an image you've just spent an hour perfecting.

Frequently asked questions

What is the best prompt structure for editing images with Nano Banana?

Name the exact subject, give one action verb, point to the specific target area, and state what must stay unchanged. One instruction per prompt works far better than combining several edits into a single request.

Why does Nano Banana change parts of my image I didn't ask it to?

This usually happens when the prompt is vague or covers multiple edits at once. Adding an explicit constraint, such as "keep the background and lighting unchanged," stops it from touching areas you didn't mention.

Can Nano Banana keep a face consistent across multiple edits?

Mostly, but it drifts after three or four rounds of edits on the same image, especially around facial features. Check against your original source image regularly and start a fresh pass from the original rather than editing an edit repeatedly.

Is Nano Banana or Nano Banana Pro better for detailed image editing?

Nano Banana Pro handles fine text, logos and multi-step compositing better, but it comes with stricter usage quotas depending on your Gemini plan, so check your plan's limits before building a workflow around it.