Photos and video
How to Use Gemini to Edit Photos
Photo editing is the single largest cluster of Gemini demand, and Gemini genuinely does it: you upload an image and describe the change in words. Two things are worth knowing before you start, because they catch people out: editing has a higher age gate than generating, and the features people most often ask for, restoring an old photo and upscaling, are not what Gemini actually documents.
GPTPrompts.AI Editorial
Hands-on with Gemini; every method here is one we use · Last updated October 2026
Affiliate link, we may earn a commission at no extra cost to you.
01 · Reality check
What Gemini actually does well here
Good at
- Local edits: changing one part of an image while leaving the rest
- Style transfer, using a second image as the reference
- Background and wardrobe swaps
- Relighting, and changing the camera angle or focus
- Resizing without cropping, and keeping a character consistent across images
Not the right tool for
- Upscaling: Google documents no upscale tool in the Gemini app
- Restoring old or damaged photos, which is also not a documented feature
- Users under 18, who can generate images but not edit them
- High-resolution output on the free tier, which caps downloads at 1K
- Precise, pixel-level control that a real editor gives you
02A · Working notes
Decide whether Gemini is the right editor
Use Gemini when you want a conversational, generative change such as removing an object, changing a background, trying a visual style or exploring a wardrobe. Use a conventional editor when you need pixel-level retouching, exact masking, colour-managed production, layered files or a reversible history. Gemini’s result is a regenerated image, not a promise that every pixel outside your request will remain identical. Choose the tool based on the artifact and risk, not on the name of the feature.
Check account, age and region limits
Before planning a workflow around a feature, check Google’s current product help for your account, country, age and plan. Product availability can vary by surface and can change. Editing and generation are not necessarily subject to the same age or access rules, and Google Photos features are separate from the Gemini app. Do not promise a client that a button will appear because another user has it. Record the date and source of any access claim that matters to the job.
Use a photo you may edit
Upload an image you own or have permission to use. For a client, family member, child, employee or identifiable stranger, confirm the relevant consent and purpose before editing or sharing. Avoid unnecessary identifiers, private documents, addresses and sensitive personal details. An image can reveal more than the prompt text. Use an approved account for confidential work and do not assume that a generated output is safe to publish simply because the edit looks harmless.
Describe the edit and the invariants
A strong prompt names the exact change and the parts that must stay: subject identity, face shape, pose, clothing, camera angle, lighting direction, shadows, grain, text and background geometry. Say what must not be added. Generative editing can alter nearby pixels or identity details even when you ask for a local change. If preserving a feature matters, mention it and inspect it after every pass. ‘Make it better’ gives the model no reviewable boundary.
Use a reference image deliberately
When you want a style, colour treatment or material, a reference image can communicate more than a list of adjectives. State what the reference contributes and what it must not copy: palette and lighting, not a person’s face or a protected design. Use references you may share. Ask Gemini to keep the subject and composition from the base image while borrowing only the named visual properties. Compare the result with both images so an attractive style transfer does not quietly change the subject.
Iterate one change at a time
Save the original, make a copy and change one variable per pass. First remove the unwanted object, then adjust the background, then tune colour or light. If you ask for five changes together, you cannot identify which instruction caused an identity shift, extra object or broken edge. Keep a short change log with prompt, output and decision. Return to the original when the image drifts rather than endlessly repairing a damaged generation.
Inspect faces, hands, text and logos
These areas deserve a dedicated zoomed review. Compare eyes, teeth, skin texture, hairline, fingers, jewellery, labels, numbers and logos with the source. A natural-looking image can still contain a changed identity or an invented product detail. Ask Gemini to describe what it changed, but treat that list as a review aid, not proof. For a commercial image, a human should inspect the final file at its intended size and confirm that any visible claim is accurate.
Do not treat regeneration as restoration
An old or damaged photo may be made to look cleaner, but a generative result can invent detail rather than recover the original. Do not present a regenerated face, document or scene as historically accurate without independent evidence. Likewise, a larger-looking image is not automatically a true upscale with recovered resolution. When provenance matters, keep the original, label the edited version and use a specialist restoration workflow with an appropriate review standard.
Check resolution and export purpose
Match the downloaded result to its use: private draft, social post, web page, presentation or print. Google documents plan-dependent download limits for some Gemini image workflows, but the current limit and access should be checked in the official help for the account you use. Do not claim print readiness from a preview. Inspect dimensions, compression, crop, colour and metadata, and keep the original alongside the export. If the file is too small, use a documented production tool rather than inventing a quality guarantee.
Handle portraits and identity safely
For a real person, preserve consent, context and identity. Do not create a misleading endorsement, intimate image, impersonation or fabricated event. Tell the model what must remain recognisable when the purpose is an ordinary enhancement, and inspect whether the result changed the person’s age, body, expression or clothing in a consequential way. For a professional headshot, obtain the person’s approval before publishing. A consented upload does not automatically authorise every downstream use.
Keep edits honest in commercial work
If an image shows a product, room, person or result that customers may rely on, label conceptual or altered imagery and check applicable advertising and platform rules. Do not edit a product to hide a defect, add an unavailable feature or imply a result that cannot be delivered. Preserve a source image and an edit record. Ask for a claim ledger: what is shown, what was changed, what is verified and what a viewer could reasonably infer. Human approval belongs with the owner of the communication.
Use a correction loop
When an edit is wrong, describe the difference from the current image rather than repeating the original brief: ‘restore the original logo text; keep the new background; do not alter the face.’ If the model keeps changing a protected area, stop and use a local editor. A correction prompt should state the desired change, protected regions, new unwanted artifacts and the acceptance check. This is more reliable than adding adjectives. Keep failed outputs out of the delivery folder and label drafts clearly.
Separate Gemini from Google Photos
The Gemini app and Gemini-powered features in Google Photos have different surfaces, requirements and availability. Do not merge their instructions or describe one as the other. If a user asks about Help me edit, confirm the current Google Photos help page, device, account and region requirements. If the task is a conversational edit in Gemini, use the Gemini documentation. Product boundaries are part of the answer, especially when a reader is troubleshooting a missing control.
Create a final review checklist
Before sharing, compare the output with the original and check identity, composition, edges, text, logos, lighting, shadows, hands, metadata, crop, resolution, consent, audience and claims. Ask whether the recipient needs the original or only the edited copy. For a client, record the approver and any disclosure. For a public post, inspect the image as a stranger and decide whether the alteration could mislead. A good checklist is faster than repairing a published mistake.
Know when Gemini should stop
Stop when you need exact restoration, a legally significant document edit, a precise product representation, an unapproved person’s likeness, a confidential source or a file that must retain layered edit history. Use a specialist tool or ask the accountable professional. Gemini is useful for exploration and ordinary creative edits, but fluent image generation does not certify accuracy, consent, ownership, provenance or commercial compliance. Preserve the original and keep the human decision visible.
Worked example: edit a product portrait safely
Suppose a small business wants to remove a distracting background from a product photo and create a warmer web version. Keep the original file, confirm the business owns the image and decide whether the product must remain an exact representation. Upload a copy and ask Gemini to change only the background and overall colour temperature, while preserving the product shape, label text, dimensions implied by the photograph, shadows and camera angle. Review the result at full size for changed lettering, edges, reflections and added objects. If the model alters the product or invents a feature, return to the original and use a conventional editor or a documented production workflow. Before publishing, compare the image with the source and create a short claim ledger: what is shown, what was changed, what is verified and what a customer might infer. If the background implies a location, season or use case that is not real, change it or disclose the conceptual treatment. Keep the edited file labelled, preserve consent for any person shown and ask the communication owner to approve the final crop and copy. The conversational prompt saved time, but it did not verify the product, grant rights, preserve provenance or approve the advertisement.
Final photo-edit handoff checklist
Keep the original beside the edited file and record the prompt, date, account surface and approver when the work matters. Check identity, faces, hands, text, logos, edges, lighting, shadows, crop, dimensions, compression, colour and metadata. Confirm the image is suitable for its intended screen or print size. Check consent, ownership, privacy and whether the edit could mislead a viewer. Distinguish Gemini from Google Photos features and verify current access claims in the relevant official help. Do not describe a regenerated image as a restoration, exact product photograph or documentary record without evidence. If a reviewer cannot tell what changed and why, the handoff is not ready.
Keep versions and decisions understandable
Name the original, working copy, approved edit and delivery export clearly. Store the prompt and review note with the version when another person may need to understand the change. Do not overwrite the source or leave a client guessing which image is authentic. If a later request reverses an edit, return to the original rather than stacking generations indefinitely. This keeps the image’s history legible and makes it easier to correct a changed face, label or background. A conversational tool is fast, but a controlled file workflow is what makes the result usable after the chat is closed.
Review the result at the intended distance
A photo can pass a close-up review and fail when it appears as a thumbnail, product card, profile image or printed page. Check both views. At small size, a changed silhouette, missing logo stroke or artificial shadow may be hard to notice; at large size, skin texture, text edges and repeated patterns reveal the generation. Compare the output with the original in the context where a viewer will see it, not only inside the Gemini preview. If the image will support a claim, inspect the copy beside it as well. A believable visual can still make an unsupported promise when the caption supplies a meaning the source never had. Approve the image and the surrounding words together.
02 · The method
Step by step
- 1
Upload and describe the change
Attach the photo and say what should change in plain language. You can also upload several images and ask for something new built from them.
- 2
Be specific about what must not change
Name the parts to leave alone. Because the model regenerates the image, saying what to preserve matters as much as saying what to alter.
- 3
Use a reference image for style
Style transfer works from a second uploaded image, which is far more reliable than describing an aesthetic in words.
- 4
Iterate rather than restart
Refine by describing the difference from the current result. Starting over from the original prompt loses the progress you made.
- 5
Try Redo with Pro if you have a plan
Paid plans expose Nano Banana Pro through the three-dot menu, which is worth using on an image you have nearly got right.
- 6
Check the resolution before you rely on it
Free downloads are capped at 1K. If you need the image for print or a large screen, the 2K download on a Google AI plan is the relevant difference.
03 · Use this now
Copy-paste prompt for a photo edit
Here is my photo. Make exactly this change: [describe the change precisely]. Keep everything else identical, especially [name the parts that must not change]. Match the existing lighting direction, shadows, colour temperature and grain so the edit is not obvious. Do not add any people, text, objects or watermarks I have not asked for. If you cannot make this change without altering the rest of the image, tell me instead of guessing. After the edit, list what you changed so I know what to check.
04 · Avoid these
Common mistakes
- Expecting an upscale or photo-restore feature. Google documents neither in the Gemini app.
- Assuming under-18 accounts can edit. Generating is 13+, but editing is 18+.
- Downloading on the free tier and finding 1K is not enough for print.
- Not naming what should stay the same, so the model changes more than you wanted.
- Confusing the Gemini app with Google Photos features, which have their own separate requirements.
05 · Questions
Frequently asked questions
Can Gemini edit photos for free?
Yes. Google documents image generation and editing as available on the free tier, running on Nano Banana 2. What the free tier limits is download resolution, capped at 1K versus 2K on a Google AI plan, and access to the Nano Banana Pro redo option, which is for paid plans.
Can Gemini restore an old photo or upscale an image?
Google documents neither as a Gemini app feature, so we do not claim they work. You can upload an old photo and describe improvements, but the result is a regenerated image rather than a true restoration. Zoom Enhance, the closest thing to upscaling in Google's products, is a Google Photos feature limited to certain Pixel phones, not a Gemini app feature.
Why can't I edit images in Gemini?
The most likely reason is the age gate. Google states that editing images is not available to users under 18, even though generating images is available from 13. Separately, the Google Photos connector is limited to 18+, US only, personal accounts, and is not available in some US regions.
What is Nano Banana 2?
It is the model behind Gemini's image generation and editing. There are variants: Nano Banana 2 Lite when you are on the Flash-Lite model, Nano Banana 2 on Flash or Pro, and Nano Banana Pro, which paid plans can invoke through the Redo with Pro option in the three-dot menu.
Can Gemini edit photos in Google Photos?
Google Photos has its own Gemini-powered editing, called Help me edit, which is separate from the Gemini app and has its own requirements: 18+, personal accounts only, Face Groups turned on, and it is not available in all regions. Ask Photos, the conversational version, is mobile only and not on the web.