Tutorials Video

Filling, Extending and Removing Parts of a Photo with AI

By Infinite Creation · Beginner · ~12 min

A 1960s mid-century comic book style editorial illustration showing a photographer with a stylus and tablet filling a gap in a harbour photograph with sea and sky, while a colleague holds up the untouched original print to compare.

Overview

Generative fill edits a selected part of a photo by describing what should be there instead. Removal and extension are the same operation: the software fills the selection with new content that matches its surroundings. This guide is for photographers, designers and creators who want to remove a distraction, widen a background or repair a damaged area. It covers what the free open source option needs, how to work without destroying the original, and where this kind of edit is not acceptable. We did not test any tool on this page. Each statement about a tool comes from the maker's own page as read on 8 October 2026, and the Sources section says what each page established.

What You Need

The original photo, saved as a lossless file where possible
An editor with a generative tool, or Krita with the AI Diffusion plug-in
For the local plug-in, an NVIDIA graphics card meeting the project's recommendation
A decision about whether this photo is a record of events or an image you are free to compose

Steps

Decide whether the edit is acceptable

Changing what a photograph shows is acceptable in art, advertising and personal work, if the audience is not misled. It is not acceptable in news, documentary and evidence, where the photograph claims to show what happened. Decide before you open the file.

Work on a copy with the edit on its own layer

Duplicate the photo. Put the generated content on a separate layer or in a smart object so you hide it, compare it with the original, or redo it. A flattened edit cannot be undone.

Choose a local or cloud tool

Krita AI Diffusion is an open source plug-in under the GPL-3.0 licence. Its project page describes it as local, open and free, recommends a graphics card with at least 6 GB of NVIDIA memory, and warns generation otherwise takes long or fails. It also offers cloud generation. Adobe's Photoshop Generative Fill is a commercial cloud tool, and we its page was unreachable, so this guide makes no claim about it.

Select generously and prompt plainly

Select a little more than the object you want removed, including its shadow. For removal, leave the prompt empty or describe the background, for example grass or sky. For extension, describe what continues. Generate several variations and compare them.

Inspect at full size

Zoom to 100 percent. Check hands, edges, repeating patterns, text and reflections, where generated content breaks. A familiar landmark or a face the viewer knows well exposes errors fastest. Fix by reselecting, or paint over by hand.

Check model licences and label the edit

The Krita project page makes no commercial-use statement for models, so read the licence of each model you load before commercial work. Disclose the generated edit where your audience would care. The labelling page covers platform rules.

Pro Tips

Small selections work better than large ones. Remove one object at a time.
The site's own background remover uses colour tolerance and is not AI. Use it for flat backgrounds and compare.
Save the prompt and settings with the project so you reproduce the edit.

Knowledge Base

Background and Detail

The tool does not know what is true about your scene. It only knows what usually appears next to the pixels you left.

Why generated fill looks right and is wrong

A fill model learned from many images what skies, grass and brick walls look like. It produces content that is statistically plausible, and has no knowledge of your actual street, building or person. When you remove a sign from a known place, the result looks natural and the place is subtly wrong. That distinction is invisible to the model and to most viewers, and it matters wherever a photo is used as evidence.

Local control versus convenience

A local plug-in keeps the photo on your computer and costs only the time and a capable graphics card. A cloud tool is easier and needs an upload. Choose by how sensitive the photo is and what hardware you have.

Where This Fits

Generative fill edits a photograph after exposure, focus and composition are fixed, which Still Photography Fundamentals covers from the start.

FAQ

Q: Is Krita AI Diffusion free?
A: Its page lists the project under GPL-3.0, calls it free and open, and offers an optional cloud service. Terms for the cloud service were not stated on the page we read.

Q: Are generated fills allowed in commercial work?
A: It depends on the model licence and the tool's terms. Read both.

Q: Will anyone know?
A: Some tools embed provenance metadata. Whether yours does, and whether a platform keeps it, varies. Label it yourself.

Vocabulary

Generative fill An editing operation that replaces a selected area with new content created by an AI model from a text prompt and the surrounding picture. The content is invented, not recovered.

Inpainting Filling a masked region of an image so it blends with its surroundings. Removal and repair use inpainting. Outpainting is the same idea applied beyond the image edge to widen the frame.

Smart object A layer in image editors that keeps the original data intact so edits stay reversible. Placing generated content in one lets you compare it with the source at any time.

Sources

The Krita AI Diffusion project page was read on 8 October 2026. No image was generated. Checked 8 October 2026.

Krita AI Diffusion, project page. States the GPL-3.0 licence, local operation, a recommendation of at least 6 GB of NVIDIA graphics memory, the warning about slow or failed generation, and the optional cloud service.

Adobe's Generative Fill page returned an access error and is not described here. No source was read on commercial terms of any model. Advice on editorial and evidential use is method guidance and was not checked against a press photography code.