outpainting

basicwidespread

Outpainting is an AI image editing technique that extends an existing image beyond its edges and invents the missing environment in a style-consistent way.

In plain terms

Imagine taking a painting out of its frame, pasting it onto a larger canvas, and asking a skilled painter to continue painting beyond the edges. The painter examines the colors, lights, and brushwork, then continues the landscape as if it had always been there. Outpainting does exactly this, only the painter is an image-generating model.

It is good for obtaining a wider image from a tightly cropped photo. A vertical image becomes horizontal, a cropped shoulder or building roof 'grows back', and a background is added around product photos so text can fit on them.

It is important that the model does not know what was actually outside the image. It does not restore, but invents a plausible continuation. The original part of the image remains essentially unchanged in most programs (minor differences may occur during blending or full image recalculation), while the new strip is an invention.

How it works

You specify where and how much the image should grow: the program enlarges the "canvas," and an empty border is added around the original image.
The model starts from the existing image: it takes into account the subject, colors, light direction, perspective, and style.
If you wish, you can specify in text what should appear in the new section (for example, "sunset at the beach"). If you write nothing, the model continues the scene on its own.
The image generation model (typically a diffusion model) fills the empty strip step by step, starting from noise, so that it fits the adjacent parts of the image.
The program blurs the seam, and usually offers several variants from which you can pick the best one or ask for a new one.

An everyday example

You took a portrait photo of your family on holiday, but the photo book requires a landscape format. With the photo editor's 'extend' button, the program paints the continuation of the beach onto both sides, and the image fills the page. Of course, those palm trees never stood there.

Who is affected

marketerFrom a single photo, you can create vertical, square, and wide advertising formats in minutes without reshooting. → Marketing

entrepreneur, executiveIt makes the production of product photos, banners, and webshop images cheaper and faster, but you are responsible for misleading images.

parent, student, teacherA good playground for creative experimentation, and also for learning that even the edge of an image can be machine-generated. → Education & Family

everyoneWith images seen in news and social media, you must expect that part of the environment was added by AI.

A common misconception

MythOutpainting shows what was actually outside the photo.

In realityThe model does not see beyond the image; it only generates a continuation that seems probable based on learned patterns. The new part is therefore an illustration, not proof of what was there.

Limits and risks

The painted-in part is fabricated, so it is misleading in news photos, documents, or evidence, and is not suitable for such purposes.
Tell-tale errors occur: distorted hands and faces, nonsensical text, broken perspective, repeating patterns, or visible seams. These are rarer in newer models but have not disappeared.
The larger the area you request, the more the result diverges from the original image's mood and logic.
Extending someone else's photo or artwork raises copyright and personality rights issues. According to the EU AI Act, from August 2026, AI-generated images that appear real and are misleading (deepfakes) must be labeled when published.

Similar, but not the same

inpainting (image completion) – Inpainting replaces or fills a selected area inside an image, while outpainting extends the image beyond its borders.
scaling up – Upscaling makes the same image crop sharper and higher-resolution; it does not add any new surroundings.
text-to-image generation – There, the full image is created from scratch based on a description, while outpainting starts from an existing image and preserves it.

Related concepts

Broader

generative AI, image generation, AI-based image editing

Narrower

generative expansion, aspect ratio conversion

On the edge

inpainting, diffusion model, scaling up, deepfake, synthetic media

A brief history

Researchers had experimented with extending images beyond their edges before, but the term and the feature became widely known in 2022. That is when OpenAI made it available in its DALL·E 2 image generator under the name Outpainting, and similar tools built on the open Stable Diffusion model also appeared. In 2023, Adobe built it into Photoshop as well: first Generative Fill, then Generative Expand, designed specifically for extending images.

Where things stand

By the end of 2026, image expansion is a standard one-button feature in popular image editors, design programs and many phone gallery apps. Similar solutions have appeared for video too, though quality there is still more uneven.

Status: 2026 November

Also known as

outpainting · out-painting · image extension · image augmentation · AI-powered image expansion · generative expansion · generative expand · image extension

Sources

OpenAI: DALL·E – Introducing Outpainting (2022)

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