AI photo editing: retouch, extend and enhance images with AI

This article is an overview based on vendor information, documentation and publicly available reporting, not a scored test of our own. Prices and features change quickly; the vendor's own terms always take precedence. We only score what we have used ourselves for weeks: those tools are listed under AI tools compared.

What AI photo editing can do today

Most of the jobs that used to take hours in Photoshop now take AI seconds: removing objects and people, cutting out the background, extending an image beyond its edges, upscaling a small file, removing noise, repairing old photos. New since 2025 is editing by sentence: "make the sky look like evening", "put a jacket on him", "place the product on a wooden table".

There are two routes. The first is a chat assistant that you give a photo and an instruction. The second is an image editor in which you mark the area yourself. Chat is more convenient, the editor more precise. If you want to create images from scratch, the tools for that are in our article Creating AI images for free.

By instruction: Gemini, ChatGPT and Le Chat

The breakthrough came with Google's image model, which became known by the nickname "Nano Banana". Since February 2026 Nano Banana 2 has been the default model in the Gemini app, including on the free plan; the stronger Nano Banana Pro comes with the Google AI Plus subscription and above. Every image carries the invisible SynthID watermark and provenance data under the C2PA standard, and on the cheaper tiers a visible Gemini symbol as well. Google does not give a fixed number of free edits. More on the assistant itself in our review of Gemini.

ChatGPT also edits photos on the free plan, since April 2026 with a fast mode; with the update of 8 September 2026, faces and details stay more stable across several editing steps, according to OpenAI. Mistral's Le Chat has used FLUX Kontext for editing since July 2025, a model from Black Forest Labs in Freiburg im Breisgau.

One piece of advice for all three: say exactly what should stay the same. "Change only the background; face, clothing and pose stay unchanged" stops the AI from quietly making someone look younger or rewriting a sign. Always compare the result with the original.

In the editor: Photoshop, Firefly and Lightroom

Adobe has consistently rebuilt its programs around AI. In Photoshop, Generative Fill fills, replaces and extends selected areas; since October 2025 you can choose Nano Banana and FLUX Kontext there alongside Adobe's own model, in which case an edit costs ten credits. Since March 2026 the Firefly Image Editor has brought fill, remove, expand, upscale and background removal together in the browser.

Adobe Firefly leads our AI image generators category, above all because of legal certainty: Adobe says it does not train on its customers' content but on licensed material. According to the pricing page, Firefly Standard costs 10.98 euros a month with 2,000 credits and Pro 21.99 euros with 4,000. For photographers, Lightroom is the obvious place: Generative Remove and AI noise reduction have been standard there since 2024.

Free and without signing up

For individual tasks, web services without an account are enough. remove.bg cuts out subjects for free, but only at a low resolution of around 0.25 megapixels; full size costs credits. Cleanup.pictures removes objects without signing up but exports only in 720p for free. Pixlr opens without an account, but its AI features use up a limited balance.

If you would rather have a full program: Affinity has been free since October 2025, but its AI features require a paid Canva subscription. GIMP 3 is free but has no generative AI built in. For upscaling small images there is Upscayl, an open-source program that runs on your own computer and uploads nothing.

Stable Diffusion offers the most control: with interfaces such as ComfyUI you can mark areas and refill them, with no quota and without a photo ever leaving your computer. The price is the learning curve and a powerful graphics card.

On your phone and your PC

Often the best tool is already installed. On the iPhone, Clean Up in the Photos app has removed distracting objects in German too since 31 March 2025, from the iPhone 15 Pro onwards. Samsung devices offer photo editing with Galaxy AI, which remains free and marks edited images with a visible symbol. On Android, Google Photos has also let users in Germany say by voice or text what should change since summer 2026.

On Windows, the Photos app erases objects with Generative Erase; Generative Fill in Paint is reserved for Copilot+ PCs. That is usually enough for snapshots, but for product photos and print it is worth taking the detour through a proper image editor.

Law: labelling, image rights, passport photos

Since 2 August 2026 the transparency obligations of Article 50 of the AI Act have applied. Providers must label AI images in a machine-readable way, and anyone who publishes a realistic image of a person, place or event that looks genuine but has been manipulated must disclose it as a deepfake. Editing that merely assists and does not substantially change the content is exempt: a removed rubbish bin is not a deepfake, a person montaged into the picture is.

Separately, image rights apply: under section 22 of Germany's Art Copyright Act, publishing photos of other people generally requires their consent, edited or not. Even uploading a portrait to an AI service counts as processing personal data.

For passport photos the question no longer arises: since 1 May 2025 German authorities accept them only in digital form, taken by certified photographers or at the office and transmitted securely. A phone photo prettified with AI cannot be used.

Where your photos go

If you upload photos to an AI service, you should know what happens to them. With ChatGPT, content from personal accounts, images included, may be used for training unless you switch off "Improve the model for everyone". With Gemini, if activity is switched on, a sample may be reviewed by humans; Google keeps that content for up to three years. Adobe rules out training Firefly on customer content.

For family photos, pictures of children and anything work-related, a tool on your own device is therefore the better choice, or a plan that excludes training by contract. What companies need to watch out for is in our article on the GDPR and the AI Act.

Which tool for which task

For quick changes by sentence, the Gemini app is currently the easiest place to start, with ChatGPT as the alternative of similar quality. For precise retouching, product photos and anything that will be published, there is hardly a way around Photoshop or the Firefly Image Editor. For cutting out or erasing individual objects, your phone's gallery or a web service will do.

If you want full control and no cost per image, run Stable Diffusion locally. And if you want to turn an edited photo into a short clip, you will find the tools in our article Image to video with AI.

In short: for changes by sentence, Gemini (Nano Banana) or ChatGPT; for precise retouching, Photoshop and Firefly; for small jobs, your phone's gallery or services such as remove.bg. Local and without uploading: Stable Diffusion and Upscayl. Label realistic manipulations of real people as deepfakes, and upload photos of others only with their consent.