AI image editing

basicwidespread

Transforming an existing photo or image with artificial intelligence: you specify what should change with words or selection, and the program realistically redraws the image.

In plain terms

AI image editing is like having a skilled retoucher sitting next to you; you only need to say: 'remove the trash can from the background', 'make it sunset', 'change the sweater to blue'. You do not have to struggle with brushes, layers, and masks; the machine understands the request and does it.

In a traditional image editor, you adjust every pixel, which requires expertise. Here, the program does not copy existing image details but invents new ones: it draws what could have been behind the erased object or how the landscape would continue beyond the edge of the image.

It differs from image generation in that it does not start from a blank page. Your photo is the starting point, and the goal is that only what you requested changes, while the rest remains untouched.

No separate program is needed anymore: it is already in phone galleries, chat AI assistants, and professional editors.

How it works

You provide the image and the request: either describe in text what you want, or use your finger or mouse to select the part to modify (or both).
The model 'looks at' the image: it recognizes objects, people, lights, shadows, and perspective, so it knows where things are and what belongs together.
The system regenerates the area to be modified. Many systems start from noise and refine it step by step into a detail that fits your request and context; others build the new image sequentially from small fragments.
In selective editing, the new part is inserted into the original: it adjusts colors, lighting, and sharpness so the seam is invisible. Tools working on text prompts, however, often redraw the entire image.
If you do not like the result, you request another version or refine the prompt. Editing thus proceeds conversationally, in multiple rounds.

An everyday example

In a family photo by Lake Balaton, everyone is smiling, except a stranger walks into the edge of the frame. In your phone gallery, you circle them, press the delete button, and within seconds water and reeds appear in their place, as if they were never there.

Who is affected

everyoneIt is in your phone, and with photos seen online, you can increasingly not be sure whether they show what actually happened.

marketerYou can create product photos, backgrounds, and advertising variations in minutes without a photoshoot or graphic designer, but you are responsible for misleading depictions. → Marketing

entrepreneur, executiveE-commerce and real estate photos can be cheaply improved, but if the image looks better than reality, it leads to complaints and loss of trust.

parent, student, teacherChildren can create humiliating or false images of each other with a few taps, so this must be discussed at home and in school. → Education & Family

A common misconception

MythThe AI modifies only the specific detail I requested; the rest of the image remains identical pixel for pixel.

In realityMany tools redraw a large portion or the entire image, so facial features, text, or background details can change imperceptibly. Always compare important images with the original.

Limits and risks

The machine does not know what was actually behind the deleted object: it invents it, so the edited image is not evidence that something looked a certain way.
Errors are common with faces, hands, text, and fine patterns; during multiple successive edits, a person's face can gradually "drift" from the original.
Transforming another person's photo without their permission can be illegal, and deceptive or humiliating forgeries can cause serious harm.
Photos uploaded to an online service go to the provider's server, so upload ID cards, children's photos, or confidential company images only with caution.

Similar, but not the same

Image generation – Image generation creates a new image from a text description out of nothing, while image editing transforms an existing image.
Deepfake – A deepfake is a realistic-looking image, video, or audio created or manipulated with AI, usually featuring a real person, which can be used for deception. AI image editing itself is the technique, mostly used for harmless corrections.
Traditional photo retouching – In retouching, a person works by hand from tool to tool on existing pixels, while AI generates new image content based on an instruction.

Related concepts

Broader

Generative AI, Generative media, Computer vision

Narrower

Image inpainting, Image extension (outpainting), Background removal, Scaling up, Style transfer

On the edge

Image generation, Diffusion model, Deepfake, AI watermark, AI video editing

A brief history

Machine filling-in of missing image regions (inpainting) is an old research area, but editing guided by text instructions became practical with diffusion models: the InstructPix2Pix paper appeared at the end of 2022. In 2023, Adobe built generative fill into Photoshop, putting the technique in the hands of the masses. In 2025, the image-editing features of conversational AI assistants (such as ChatGPT and Gemini) made editing by conversation an everyday thing.

Where things stand

By the end of 2026, conversational editing has become a standard function in phone galleries, chat AI assistants, and professional editing software. Attention has shifted to labeling edited images and verifying their origin.

Status: 2026 November

Also known as

AI image editing · AI image editing · AI image editing · AI photo editing · Generative image editing · AI image editing · AI photo editing · generative fill

Sources

Brooks, Holynski, Efros: InstructPix2Pix – Learning to Follow Image Editing Instructions (arXiv, 2022)

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.