The short answer
To edit a photo with AI, upload it to a tool that takes plain-language edits and describe one change at a time. In the Gemini app, open the sidebar, choose Library, pick or upload the image, click Chat under it and type the edit. On Android, Google Photos offers Edit, then Ask, for conversational edits and Magic Eraser for removing objects. For commercial design work, Adobe Firefly's own models are trained on licensed content, and Canva's Magic Edit works inside your designs.
What you need first
- The original file. Always keep an untouched copy. Edit a duplicate.
- One clear change in mind. "Remove the bin behind the table" works far better than "make this look professional".
- The rights to the photo, and permission from anyone recognisable in it if the result will be published.
- The right tool for the job. Quick personal fixes: your phone's photo app. Bigger changes from a description: Gemini. Brand and commercial work: Adobe Firefly or Canva.
How to edit a photo in the Gemini app, step by step
- Go to gemini.google.com on a computer and sign in.
- Click Open Sidebar, then Library.
- Click the image you want to edit, or upload one from your device.
- Under the image, click Chat.
- Type your edit in the text box at the bottom, for example "remove the person in the red jacket on the left and fill the background naturally".
- Click Submit, check the result and ask for the next change in the same chat.
Which image model you get depends on your settings and plan, according to Google's help centre. With the Flash-Lite model you get Nano Banana 2 Lite, built for speed rather than multiple edits. With Flash or Pro you get Nano Banana 2, which accepts several reference images. Nano Banana Pro needs a Google AI plan and the Pro model: click More on an image, then Redo with Pro. Downloads are 2K with a Google AI plan and 1K without. Editing images is not available to users under 18.
How to edit a photo in Google Photos on Android
For everyday fixes on your phone, Google Photos documents these tools:
- Conversational edits: open the photo, tap Edit, then Ask, tap Continue, type or say a request such as "remove the glare" and tap Save.
- Removing objects: tap Edit, then Search, type "Magic Eraser", circle or brush the object and tap Erase. Tap Done, then Save as copy to keep your original.
- Quick improvements: tap Edit, then AI Enhance, and choose one of the results.
- Reframing: tap Edit, then Crop, then Auto frame, which can fill in or expand parts of the picture.
Most of these need at least 3 GB of RAM and Android 8.0 or later, some are Pixel-only, and some tools must be downloaded first over Wi-Fi or mobile data. Google Photos also lets you check AI info for edited photos.
A worked example: a product photo for an online shop
Say you sell handmade candles and your best photo has a phone charger cable in the corner and a dull, grey cast. In Gemini, open the photo from Library and click Chat. First request: "Remove the white cable in the bottom right corner." Check it, then: "Make the lighting warmer, like late-afternoon sun from the left, without changing the candle's colour or label." Then: "Extend the background to a square format with the same wooden table." Three small requests beat one long one, because you can see which change went wrong. Download the result, compare it side by side with the original and make sure the product itself looks exactly as a customer will receive it. If the photo is for a paid advert rather than your own shop listing, consider making the final version in Adobe Firefly with Adobe's own models, for the reason in the next section.
Commercial use and honest editing
- Adobe Firefly says it does not train Firefly or partner models on users' content, and its own Firefly models are trained on licensed Adobe Stock and public domain content. Its commercially safe promise covers Adobe's own models only. For partner models such as GPT Image or Nano Banana, Adobe says it is up to you to decide whether they suit a commercial project. Firefly outputs carry Content Credentials showing AI was used.
- Canva includes Magic Edit, Magic Eraser, Magic Expand, Magic Grab and Background Remover, so you can fix a photo inside the design it is going into. How much AI use each plan allows is set out on Canva's pricing page. Canva's trust page says content on Business, Teams and Enterprise accounts is not used to improve its AI features.
- Do not mislead. Removing a stray cable is fine. Changing what a product looks like, adding people who were not there, or altering a news, property or insurance photo is not. Check the rules of the platform where you publish, because some ask you to disclose AI edits.
- People need consent. Do not edit a recognisable person into a situation they were not in, and do not publish edited photos of colleagues or customers without their permission.
Common mistakes and how to fix them
- Several changes in one request. The tool gets one wrong and you cannot tell which. Fix: one edit per message.
- The product changes shape or colour. Generative edits can redraw the whole image. Fix: say what must not change ("keep the label, logo and colours exactly the same") and compare with the original before publishing.
- Text and faces come out wrong. Fix: add text afterwards in a design tool, and avoid regenerating faces.
- Overwriting the original. Fix: use Save as copy in Google Photos and keep originals in a separate folder.
- Low-resolution downloads. Gemini downloads are 1K without a Google AI plan. Fix: use a paid plan or a design tool for print-size work.
- Uploading private photos without checking settings. Fix: check each tool's privacy settings first; for Gemini, see Keep Activity at myactivity.google.com/product/gemini.
Which tool should you use?
- Google Photos for quick fixes to personal photos on Android.
- Gemini for bigger changes described in words, with no design skills needed.
- Adobe Firefly for commercial and brand work, when you want Adobe's licensed-content models and Content Credentials.
- Canva when the photo is going into a social post, presentation or flyer you are already designing.
For more, read the best AI for images, the best AI for designers, Midjourney vs Adobe Firefly and Canva AI alternatives.
Prompts to copy
Remove a distraction
Remove [object] from [location in the photo] and fill the area so it matches the surrounding [background]. Do not change anything else in the image.
You get: A clean photo with only the distraction gone
Fix the light
Make the lighting [warmer / brighter / softer], like [light source and time of day] from the [direction]. Keep the colours of [product or person] exactly as they are.
You get: Better light without changing what matters
New background for a product
Place this [product] on [new surface] in [setting], with [light]. Keep the product's shape, label, logo and colours exactly the same, and keep the shadow realistic.
You get: A product shot in a new scene that still shows the real product
Extend to a new format
Extend this photo to a [square / 9:16 vertical / 16:9 wide] format by continuing the [background] naturally. Do not crop or move the main subject.
You get: A photo that fits the format you need without cropping the subject
Restore an old photo
Repair the scratches, dust and fading in this old photo. Keep faces, clothing and the original composition unchanged. Do not add colour.
You get: A cleaner version of an old photo that stays faithful to it
Honesty check before publishing
I edited a photo for [use, such as a product listing or advert]. The edits were: [list]. Tell me which edits could mislead a viewer, which might need disclosure, and what I should check against the original before publishing.
You get: A quick check that your edits stay honest