Gemini Image Background Change: Prompts, App Steps, PNG Limits

Yes, Gemini can change the background of a photo you upload. Open the Gemini app on the web, iPhone or Android, attach the photo, and tell it to change only the background to the scene you want. Then check the face, hair and edges before you download. Editing an uploaded photo needs a signed-in personal account and age 18 or over. Without a Google AI plan you download at 1K; with one, at 2K.

Two limits shape everything else. Gemini returns a flat image, not a transparent PNG, so a "remove the background" request gives you a new background (white, a color, a scene), never true transparency. And because the whole picture is regenerated, faces and small details can shift, so the check before downloading matters. For hundreds of photos, the Gemini API costs $0.067 per 1K image with Nano Banana 2 as of October 4, 2026.

How to change a photo background in the Gemini app, step by step

The flow is the same on gemini.google.com and in the iPhone and Android apps. Google documents no difference between them for image editing.

1. Sign in with a personal Google account. Per Gemini Apps Help, editing images isn't available to users under 18. Image features also only work in the app's supported countries and languages, and work or school accounts may have other restrictions.
2. Attach the photo. Upload it from your device in the prompt box. To edit an image Gemini made earlier, open Library on the web, pick the image, and choose Chat.
3. Type the change. Name the subject, say that only the background changes, and describe the new background. Copyable prompts are in the next section.
4. Check the result at full size. Look at the face, hairline, hands, and any logo or label text. If something drifted, reply in the same chat with one correction, such as "Keep everything the same, but restore the original face."
5. Try again if needed. Ask for another version, or start over from the original photo if several rounds of edits have started to degrade it.
6. Download the full-size image. Gemini saves at 1K without a Google AI plan and at 2K with one.

The model behind this is Nano Banana 2 for everyone, as of October 5, 2026. If the chat is set to Flash-Lite, Gemini uses Nano Banana 2 Lite, which Google says "isn't designed for multiple image references or for multiple edits," so switch to Fast, Thinking or Pro for background work. Subscribers on Google AI Plus, Pro or Ultra can also regenerate a result with Nano Banana Pro via More > Redo with Pro. That option stops working once you hit the daily Nano Banana 2 quota. If the Pro option is missing, Nano Banana Pro not working in Gemini walks through the causes.

Every image Gemini makes carries an invisible SynthID watermark and Content Credentials, and those always stay. The visible watermark is now a setting you control, with or without a plan: in the Gemini app, go to Settings > Media Watermark and switch it on or off (Google's watermark settings help). The exception is India, South Korea and Vietnam, where the switch only appears with an AI Ultra subscription and visible watermarks apply otherwise.

Two similarly named features do different jobs. Gemini in Google Meet creates video-call backgrounds, and the Gemini app's own theme settings change how the chat looks. Neither edits a photo you already have.

Prompts that change only the background and keep the subject

Google's image-generation guide has a template built for this, which it calls inpainting with semantic masking. You describe the area to change in words instead of drawing a mask:

"Using the provided image, change only the [specific element] to [new element/description]. Keep everything else in the image exactly the same, preserving the original style, lighting, and composition."

Two more tips from the same guide matter for backgrounds. First, describe what you want rather than what you don't: "an empty, deserted street with no signs of traffic" works better than "no cars." Second, refine in small follow-ups, such as "That's great, but can you make the lighting a bit warmer?"

The prompts below adapt that template to common jobs. They are starting points, not guaranteed recipes, and results vary with the photo. Swap in your own subject and scene.

Portrait in a new outdoor scene

Using the provided photo, change only the background to a quiet city park in late-afternoon light, softly out of focus. Keep the person exactly the same: face, hair, expression, clothing and pose. Match the light on the person to the new scene.

Product on a white studio background

Using the provided photo of this product, change only the background to a seamless pure white studio backdrop with a soft, natural shadow under the product. Keep the product's shape, colors, logo and label text exactly the same.

Plain solid color, such as light gray for a profile photo

Using the provided photo, change only the background to a flat, even light gray with no texture, gradient or objects. Keep the person and the edges of their hair exactly as they are.

Blur the existing background instead of replacing it

Using the provided photo, keep the same scene but make the background look like it was shot with a wide aperture: a smooth, soft blur behind the subject. The subject stays sharp and unchanged.

Clean up clutter behind a pet or object

Using the provided photo of my dog, change only the background to a clean, empty wooden floor and a plain off-white wall. Keep the dog exactly the same.

Make the subject blend into the new background

Run this as a follow-up when the subject looks pasted on:

Adjust only the lighting on the subject so it matches the background: light from the same direction, the same warm color, and a soft shadow on the ground beneath them. Do not change the face, clothing or pose.

A few habits make these work better. Change one thing per message: replace the background first, then fix the lighting in the next turn. Name the details that must survive (face, logo, text) instead of writing "keep everything." And when you want something gone, describe the clean result ("an empty beach") rather than listing what to remove.

Can Gemini make a transparent background? Not a real PNG

No. Google's image documentation lists no transparency or alpha-channel option for Gemini image models, and the output formats it shows are PNG and JPEG with no transparency setting. In practice, asking for a "transparent background" can return a gray-and-white checkerboard painted into the pixels. It looks transparent in the preview but turns into a solid pattern the moment you place it on another color. The same trap shows up in OpenAI's models; GPT Image transparent background: fix a baked-in checkerboard explains how to spot it.

If you need a real cutout, you have three options:

• Remove background in Google Slides, Drawings or Vids. Click the image, choose Edit image > Remove background, or go through Format > Image > Edit image > Remove background. Google's help page says it "requires an eligible Google Workspace or Google One AI Premium subscription" (Google One AI Premium is the older name for today's Google AI plans). In Vids the generative features are English-only. Reset image can't bring a removed background back, so use Undo or Version history instead. Google's instructions list the details.
• A dedicated background remover for photos, especially if you need a transparent file to reuse elsewhere.
• Gemini first, cutout second. Ask Gemini for a flat white or single-color background, then run that result through a remover. A plain, even background can make the edge easier for the remover to find.

Whichever route you take, test the file on a dark background before you send it. How to make an image transparent (and check it actually is) covers that check and when to use which kind of tool.

Why Gemini changed the face, hair or label text

Gemini redraws the whole image when it edits. So even when the background is the only thing you asked about, the subject is regenerated too. Google lists character consistency as a Nano Banana 2 feature, but users often report that a face looks slightly different after an edit. Fine hair and small print on a product are regenerated the same way, so treat the check as part of every edit.

What helps:

• Name what must not change. "Keep the person exactly the same: face, hair, expression" gives the model a clearer target than "keep the subject."
• Correct one thing in a follow-up. "Keep everything the same, but restore the original face from my photo" is more precise than rerunning the full prompt.
• Restart from the original after a few rounds. Each edit builds on the previous output, so small changes can add up. A new chat with the untouched photo gives a clean base.
• Compare side by side at full size. Small changes to eyes, teeth, logos and text are easy to miss on a phone preview.

For ID or passport-style photos, the issuing agency sets the photo rules, not Gemini. Check those rules before submitting a photo with an AI-edited background, since an edited photo may not be accepted.

When Gemini won't edit the photo: age, account, limits or people

When Gemini declines an edit or returns nothing, the cause is usually one of these:

Photos of people need extra care. Google's help pages remind users not to violate others' privacy rights, and Gemini may remove images when its systems detect a possible policy violation. Google's enforcement page names "using someone's image without their consent" as one of the things it looks for. Repeated violations can restrict your use of Google's generative AI or your Google account.

The normal use is fine: Google itself suggests using Nano Banana 2 to "picture yourself in different scenes," so changing the background behind your own portrait is exactly what the feature is for. If a legitimate edit of your own photo is still refused, rephrase the request to describe the background change plainly and try again. Don't try to get around a refusal. Google's policy treats circumventing safety filters as a violation in itself. Why Gemini restricts images of people explains where those lines sit.

Gemini app, Slides or the API: which fits your photo

Prices as of October 4, 2026, from Google's Gemini API pricing page.

Choose Gemini for a replacement scene, a background remover for a true transparent cutout, and the paid API for repeatable bulk requests. A white backdrop is still a flat background, not transparency.

Change backgrounds in bulk with the Gemini API: per-image costs

The API makes sense when you have a folder of photos and the same instruction for each. It isn't free. As of October 4, 2026, Google's pricing page lists image output as "Not available" on the free tier for every Gemini image model, and the paid tier needs billing with at least $5 prepaid. Google AI Studio's playground also asks for an API key or a Google AI plan before it will run Nano Banana 2. If you were hoping for a free route, see whether the Gemini image API has a free tier and what Google offers for Nano Banana at no cost.

The photo you send is billed as input, not output: $0.50 per million input tokens on Nano Banana 2, and about $0.0011 per input image on Nano Banana Pro. That is small next to the output price.

A worked example: 500 product photos changed to white at 1K on Nano Banana 2 cost 500 × $0.067 = $33.50 at Standard prices, or 500 × $0.034 = $17.00 through the Batch API, which targets a 24-hour turnaround. Google's $5 minimum prepayment covers about 74 Standard 1K edits (5 ÷ 0.067 = 74.6, rounded down). These figures leave out the small input and text charges.

Older tutorials point to Imagen's mask modes for background replacement. The Imagen 4 models were shut down on the Gemini API on August 17, 2026. Gemini image models take no mask parameter; the "mask" is the sentence in your prompt that says what to change.

Here is a minimal Python call with one photo in, one prompt, and one image out. It follows the request shape in Google's image-generation guide and was not executed for this page, so try it on a single test photo first. It needs the google-genai package and your key in the GEMINI_API_KEY environment variable.

import base64
from google import genai

client = genai.Client()  # reads GEMINI_API_KEY from the environment

with open("photo.jpg", "rb") as f:
    photo = base64.b64encode(f.read()).decode("utf-8")

prompt = (
    "Using the provided photo of this product, change only the background "
    "to a seamless pure white studio backdrop. Keep the product exactly the same."
)

interaction = client.interactions.create(
    model="gemini-3.1-flash-image",
    input=[
        {"type": "text", "text": prompt},
        {"type": "image", "data": photo, "mime_type": "image/jpeg"},
    ],
)

with open("photo-white-background.png", "wb") as f:
    f.write(base64.b64decode(interaction.output_image.data))

To pay for 2K or 4K instead of the 1K default, Google's guide sets response_format with an image_size of "2K" or "4K" (uppercase K). The request shape is the same for Nano Banana Pro; the Nano Banana Pro API guide has more templates. When a request returns text but no image, the generateContent API reports the reason in finishReason, for example IMAGE_SAFETY or IMAGE_OTHER. Nano Banana 2 blockReason OTHER explains what to check. Requests that fail with a 400 or 500 error aren't charged, but they still count against your quota. The Gemini API is available only in Google's listed regions, which include the United States.

If you'd rather pay per call than manage Google billing, laozhang.ai is a paid gateway that lists Nano Banana 2 at $0.055 per call (public pricing as of September 24, 2026). That is below Google's $0.067 Standard 1K price but above the $0.034 Batch price. A response that returns HTTP 200 without an image still counts as a charged call. Output size and aspect ratio can only be set on its Gemini-native route. Requests still go to Google's model, so the same content rules apply.

Disclosure: laozhang.ai is our own service. Its prices are platform pricing, not an independent comparison.

Questions about changing photo backgrounds in Gemini

Is it free to change a photo background in Gemini?

In the Gemini app, yes, within the usage limits for accounts without a plan. You get 1K downloads, and you can turn the visible watermark off under Settings > Media Watermark (outside India, South Korea and Vietnam). A Google AI plan raises the limits and the download size to 2K. The API has no free image output. Can you use Nano Banana for free? compares Google's no-cost options.

How do I change a photo background with Gemini on iPhone?

Use the same steps as on the web: open the Gemini app, attach the photo from your library, type a "change only the background" prompt, check the result, and save the full-size image. The age, sign-in and plan rules are the same on every device.

Can Gemini remove a background and leave it transparent?

No. Gemini replaces the background with something, even if that is plain white. A "transparent" result that shows a checkerboard is a painted pattern, not real transparency. Use Remove background in Google Slides, Drawings or Vids (eligible plan required) or a dedicated remover.

Why does my face look different after Gemini changed the background?

Gemini regenerates the whole image, including the person. Name the features to keep ("face, hair, expression"), fix drift with a one-line follow-up, and restart from the original photo if repeated edits have piled up changes.

Can I use Gemini to make a passport or ID photo background white?

Gemini can make the background white, but it can't make the photo meet an official standard. The agency that issues the document sets the rules for backgrounds, lighting and editing, so check them before you submit an edited photo.

Can I change the background of a photo of someone else?

Only if you have that person's permission. Google's policies prohibit using someone's image without their consent, and Gemini may remove images it flags.