How to use Photoshop Generative Fill and Generative Expand

Photoshop Generative Fill lets you select part of an image, type what you want there and choose from generated versions that try to match the photo’s lighting and perspective. Leave the prompt empty and it fills the selection from the surrounding pixels instead, which is how you use it to remove something. Its companion, Generative Expand, does the same job outside the original frame when you widen the canvas with the Crop tool.

This guide covers the steps in Photoshop on the desktop, how to get cleaner results, what each generation costs in credits and why the button is sometimes greyed out. It is based on Adobe’s own help pages, checked on 30 October 2026. Menus move between Photoshop versions, so if a step does not match your screen, update the app first.

Before you start

Adobe’s Photoshop generative AI FAQ sets out the basics:

• Who can use it: Photoshop subscribers and trialists, wherever Adobe provides its services.
• Internet: you need to be online to use these features.
• Prompt language: Adobe says prompts work in more than 100 languages.
• Any image you can open: Generative Fill and Generative Expand work on any document Photoshop can open, including Adobe Stock images and pictures made with other AI image generators.

Adobe’s troubleshooting page adds the file conditions: the image should be in RGB mode at 8 bits per channel, and you need to work on an unlocked raster layer rather than a text layer, shape or smart object.

How to use Generative Fill

The steps in Adobe’s Generative Fill help page are:

1. Make a selection. Use any selection tool, such as the Selection Brush, Lasso or Object Selection tool, around the area you want to change.
2. Open Generative Fill. Select it in the Contextual Task Bar that appears under the selection, or go to Edit > Generative Fill.
3. Write a prompt, or leave it empty. Describe what you want to appear, such as “a wooden bench”. To remove what is inside the selection, leave the box blank.
4. Select Generate. Photoshop shows variations in the Properties panel. Adobe’s tutorial on removing objects shows three per generation.
5. Pick a variation. Click each thumbnail to compare. Change the prompt and select Generate again if none of them works.

The result lands on a new Generative Layer with a mask in the shape of your selection. Your original pixels stay underneath, so hiding or deleting that layer brings the original back.

Getting a cleaner result

A few habits from Adobe’s own tutorials make a difference:

• Leave some room around the object. Adobe’s tutorial expands the selection by a set number of pixels before generating, noting that Generative Fill often gives better results with some breathing room between the selection and the object. In the Contextual Task Bar, use Modify selection > Expand Selection.
• Name the object, not the scene. Adobe’s Photoshop image generation page says Generative Fill matches the lighting, shadows and perspective of the image without needing detailed text descriptions.
• Sharpen the variation you chose. Adobe’s Enhance Detail option, in the Properties panel under Variations, makes a sharper copy of a variation. Adobe notes it only works on variations larger than 1,024 pixels.
• Mind the resolution on big selections. Adobe’s desktop generative AI FAQ says the Firefly model returns output of 1,024 × 1,024 pixels to Photoshop, so a fill across a large area of a high-resolution photo may look softer than the pixels around it.

Expanding a photo with Generative Expand

Generative Expand is built into the Crop tool. Adobe’s Generative Expand page gives these steps:

1. Select the Crop tool and drag the crop handles outward, past the edge of the image.
2. Select Generative Expand in the Contextual Task Bar.
3. Choose a Firefly model from the model picker if you want a different one. Adobe lists Firefly Fill & Expand, Firefly Image 3 and Firefly Image 1.
4. Leave the prompt blank so Photoshop continues the scene, or describe what should appear, then select Generate.
5. Choose a variation in the Properties panel.

Adobe’s expand tutorial shows a practical use: setting the crop’s aspect ratio to 4:5 to turn a landscape photo into a portrait post without cutting off either side.

What it costs in generative credits

Each generation uses Adobe’s generative credits, and how many you have depends on your plan. Adobe’s generative credits FAQ says most standard features, such as Generative Fill using Firefly models in Photoshop, use 1 credit per generation. Generative Expand is also listed at one standard credit per generation.

The plan makes a big difference:

On a newer Photoshop-only plan, 25 credits can go quickly when each attempt produces a batch of variations you may not use. If you only need to remove a small distraction, Adobe’s FAQ lists Photoshop’s Remove Tool among the features that do not currently use credits. Our guide to Adobe Firefly generative credits explains the allowances, reset dates and how to check your balance.

Why Generative Fill is greyed out

Adobe’s troubleshooting page lists the usual causes:

• No active selection. Generative Fill needs a selection on the image.
• Wrong colour mode or bit depth. Use RGB mode at 8 bits per channel, not 16 or 32.
• Wrong layer type. Select an unlocked raster layer. Rasterise a smart object first if you need to fill inside it.
• No credits left. Check your balance at account.adobe.com, wait for the monthly reset or change plans.
• A firewall, proxy or VPN in the way. Adobe says some users find these block its AI services.
• An old version or a school or work account. Update Photoshop, and note that some Enterprise and Education accounts have restricted AI access.

Saving your work

Every Generate creates a new Generative Layer. Adobe’s FAQ says to save as a PSD or another layered file type to keep those edits; if the file is already layered, a simple Save keeps them.

If you plan to publish or sell the result, read our guide on using AI images commercially first. Adobe’s Firefly FAQ says it applies Content Credentials to certain exports that use Firefly output, to show an AI tool was involved; our guide to telling whether an image is AI-generated explains how people read them.