How to use Photoshop AI in 2026: Generative Fill, Expand and Remove in 5 steps

To use Photoshop AI, open your photo in Photoshop on the desktop, make a selection, click Generative Fill in the Contextual Task Bar, type a short prompt and hit Generate. Photoshop gives you three variations on a new layer, and you pick the one that fits. Generative Expand (through the Crop tool) and the Remove tool cover most of the other everyday jobs.

That's the short version. The longer one matters because Adobe's moved the goalposts a few times since 2023: there are now partner models that cost more credits, a Reference Image option, and at least one forum report that the old "leave the prompt empty" trick doesn't work the way it used to. Adobe's own help pages blocked our requests while we researched this in late October 2026, so the steps below follow menu names reported in recent tutorials from Dupple, Photoshop Essentials and Creative Bloq. If a label looks different in your copy, that's probably why.

Before you start: what you need

You'll need a Photoshop subscription and a fairly recent version. Going by Dupple's 2026 guide, generative features arrived in version 25.0 (the 2024 release), and any Creative Cloud plan with Photoshop gets them, starting with the Photography plan at $9.99 a month. Generation runs on Adobe's servers, so you have to be online when you click Generate.

Then there's the credit system. It's confusing, and I don't think that's entirely an accident.

Every generative action spends "generative credits", which reset monthly and don't roll over. CGPress pulled Adobe's scattered numbers into one guide: new Photography and single-app subscribers get roughly 25 credits a month, while Creative Cloud Pro offers up to 4,000 for premium features. Standard Firefly generations cost about one credit each. Generative Fill with a partner model costs 10 credits, and Google's Gemini 2.5 Flash Image is listed at 20. On Adobe's community forum, one user writes that people on similar plans report getting 250 or 500 credits, which tells you how muddled this is. Your own number shows up when you click the small pie-chart icon in the Creative Cloud app.

A quick checklist before you begin:

• Photoshop on desktop, updated to the latest release from the Creative Cloud app.
• An internet connection while generating.
• A few spare credits (check your balance first).
• A copy of your original file, or at least a duplicate layer (Ctrl/Cmd + J).

The five steps: how to use Photoshop AI

These steps follow one job from start to finish: select, generate, choose, extend, clean up. Each step lists what you should see on screen, since we couldn't include our own screenshots.

Step 1: Open your image and turn on the Contextual Task Bar

Open the photo with File > Open. Most of Photoshop's AI lives in the Contextual Task Bar, a floating strip that appears under whatever you've selected. If it isn't showing, Creative Bloq's walkthrough says to switch it on under Window > Contextual Task Bar. You can drag it out of the way and pin it, though that position resets with each new document.

What you'll see: a small bar with buttons such as Select subject and Remove background.

Step 2: Select the area you want to change

Draw a selection with the Lasso, the Object Selection tool or the Selection Brush. Rough is fine. Dupple suggests leaving a small margin around an object so the fill blends better, and it points out that shape matters: a tall, narrow selection suits a standing person far better than a wide square one.

Want a new background? Use Select > Subject, then Select > Inverse, so everything except your subject is selected.

Step 3: Click Generative Fill and write your prompt

Click Generative Fill in the task bar, type what you want inside the selection, then click Generate. Keep it plain and short; Dupple recommends 5 to 12 words and skipping filler like "8K" or "photorealistic", because Photoshop already tries to match the photo's look. A few prompts you can copy:

golden retriever sitting on the grass, afternoon light from the left

light oak table, soft window light

clear blue sky

Next to the prompt box there's a model dropdown. Digital Camera World notes it reads Firefly by default and lets you switch to partner models such as Google's Nano Banana (Gemini 2.5 Flash Image) or Flux. Engadget's Adobe MAX 2025 report lists the same partners, plus Black Forest Labs' FLUX.1 Kontext. CGPress says hovering over Generate shows what the selected model will cost, which I'd do every single time before switching away from Firefly.

There's also a Reference Image option. As Fstoppers explains, you can upload a picture (a specific necklace, say) and Photoshop uses it to guide what it generates in the selection, rather than relying on words alone.

Step 4: Pick a variation and refine it

Each Generate click returns three options on a new generative layer with its own mask. Flick between them in the Properties panel. If none work, tweak the prompt and generate again; earlier results stay on their layers, so you can mask between attempts.

Adobe's latest Firefly update outputs up to 2K resolution with fewer artefacts, according to Creative Bloq's report on the update. Still, its writer cautions that "this isn't a magic switch". Zoom to 100% before you accept anything, since a fill that looks fine zoomed out probably won't hold up in print.

Step 5: Extend the frame or clean up with Generative Expand and the Remove tool

To make the canvas bigger, pick the Crop tool (C), drag the handles past the edge of the image, then click Generative Expand in the task bar. Leave the prompt blank to continue the scene, or describe what should appear ("more trees", say). Dupple's advice is to expand 20 to 30 percent at a time instead of doubling the width in one go.

For removing things, the Remove tool is usually quicker. It's nested under the Spot Healing Brush (J). Photoshop Essentials recommends setting Mode to Generative AI on, ticking Sample all layers and working on a blank layer above the photo. Paint over the object loosely, or click Find distractions and choose People or Wires and cables; Photoshop highlights its picks in magenta, you add anything it missed, then click the check mark. CGPress lists the Remove tool among the features that don't consume credits, though I couldn't confirm that on Adobe's own pages.

Example: a complete worked example

Say you've shot a coffee mug for an online shop. The desk behind it is messy, there's a charging cable in frame, and you need a wide 16:9 banner as well as the square product shot.

1. Open the photo and press Ctrl/Cmd + J to duplicate the layer. Add a blank layer on top.
2. Pick the Remove tool, make sure Generative AI is on and Sample all layers is ticked, then paint over the cable. Check the patch at 100% and touch up any smudges with a second stroke.
3. Click the mug, choose Select subject in the task bar, then Select > Inverse to select everything else.
4. Click Generative Fill, leave the model on Firefly and enter the background prompt below. Generate, then compare the three options in the Properties panel.
5. Choose the Crop tool, pick a 16:9 ratio and drag the left and right edges outward by about a quarter. Click Generative Expand with no prompt. Repeat once if you need more width.
6. Save a PSD to keep the layers, then use File > Export > Export As for the JPEG or PNG.

light oak table, plain white wall, soft window light from the left

On Firefly, that's roughly three or four standard generations, plus any retries. Switch the fill to a partner model and the same job costs at least 10 credits per click, by CGPress's figures. On a plan with about 25 a month, that adds up fast.

Common mistakes and how to fix them

The Generate button is greyed out with an empty prompt

Older tutorials, including Creative Bloq's, say an empty prompt makes Generative Fill remove the selected object. That likely no longer holds for everyone. In a thread on Adobe's community forum, one user writes that "the Generate button often stays grayed out when the prompt field is empty". The fix: use the Remove tool for removals, or describe what should be there instead ("grass and dirt path").

Selections that are too big

Huge selections tend to come back soft or inconsistent. Work in smaller chunks. Dupple also warns that very large files (100MP and up) can time out, so it suggests editing at a moderate size and upscaling afterwards with Image > Generative Upscale.

Expecting readable text, hands or a specific face

Signs and logos usually come out garbled, and hands can still look wrong, Dupple notes. Add text with the Type tool instead. And you can't prompt for a named real person, which I think is the right call.

Burning credits on the wrong model

The model dropdown remembers your last choice. If you tried Gemini once and forgot, every later fill costs 20 credits rather than one. Check the dropdown before you click.

Tips to get better results

• Mention the light that's already in the photo ("warm afternoon light") so additions match.
• Describe the camera angle for objects ("coffee cup on table, viewed from above").
• Use Reference Image when you need a specific product or accessory, not a generic one.
• For composites, try Harmonize: Digital Camera World says it sits in the Contextual Task Bar when a pixel layer (not the background) is selected, and it matches colour, light and shadow.
• Only edit photos of people with their consent. Making deceptive or intimate images of real people is harmful and illegal in many places.

One opinion, for what it's worth: I'd learn the Remove tool before Generative Fill. Most everyday edits are removals, and if CGPress is right that it doesn't spend credits, it's the cheapest AI in the app.

Next steps

Once the basics click, there's a chat-style option too. eWeek reported that Adobe's AI Assistant is in public beta for Photoshop on the web and on mobile, where you describe an edit ("remove distractions", "change the background") and it either applies it or walks you through the tools. It's not in the desktop app, as far as I can tell from the coverage we found. We also couldn't confirm what, if anything, Adobe changed in Photoshop at its October 2026 event, so check the What's New panel in your copy.

If you'd rather generate whole images from scratch, our Adobe Firefly review covers Adobe's standalone app, and our list of the best AI image generators compares it with rivals. Our AI photo prompt guide helps with wording. For cheaper clean-up tools, see the best AI object removers and AI photo retouching tools. And if you'd like a browser-based generator with more model choice, our OpenArt AI tutorial is a good next read.

FAQ

How to use AI in Photoshop?

Open an image in Photoshop desktop, select an area, then click Generative Fill in the Contextual Task Bar. Type a short prompt and click Generate. You'll get three variations on a new layer. Generative Expand (via the Crop tool) adds canvas, and the Remove tool erases objects. You'll need a subscription, credits and an internet connection.

How to use Photoshop AI generative fill?

Make a selection with the Lasso or Object Selection tool, click Generative Fill in the task bar and describe what should appear, such as "clear blue sky". Click Generate, then pick one of the three results in the Properties panel. Each result sits on its own masked layer, so you can refine or blend it without damaging the original photo.

How to use AI generative fill Photoshop?

It's the same process: select, prompt, generate, choose. Keep prompts short and describe the content, not the style. If Generate stays greyed out with an empty box, type what the background should look like. Firefly costs about one credit; partner models cost 10 or more.

How to do generative AI in Photoshop?

Photoshop's generative AI runs through Generative Fill, Generative Expand, Generative Upscale, Harmonize and the Remove tool. All but the Remove tool spend generative credits from your monthly allowance, which resets and doesn't roll over. You choose Firefly or a partner model from a dropdown next to the prompt box, and you need to be online.

How to photoshop with AI?

Start with the Remove tool for clean-up; it handles people, cables and clutter with a brush stroke. Use Generative Fill to add or replace things and Generative Expand to change the shape. On web and mobile, the AI Assistant beta takes plain-language edits.