What AI Tools Are Available in Photoshop for Creators (And Which Ones I Use)

The short version: Photoshop now packs Generative Fill, Generative Expand, Generative Remove, Neural Filters, Sky Replacement, Select Subject, and a text-to-image generator built straight onto the canvas, all running on Adobe's Firefly model. Most creators only ever touch two of them. The rest either need a bigger project to justify the click or, honestly, produce results you'll spend longer fixing than if you'd just shot it in the first place.

Where Photoshop's AI tools live

Adobe has scattered AI across nearly every panel in Photoshop now, so it's easy to miss half of what you're paying for. If you're on any current Creative Cloud plan, here's what's sitting inside the app right now:

• Generative Fill (add, remove, or replace anything inside a selection)
• Generative Expand (extend a canvas beyond its original edges)
• Generative Remove (a dedicated, faster version of Fill for deleting objects)
• Neural Filters (skin smoothing, Smart Portrait, colourise, depth blur, photo restoration)
• Select Subject and Select Sky (one-click masking)
• Sky Replacement
• Camera Raw's AI Denoise
• Text-to-image generation directly on a new layer (the on-canvas "Generate Image" option)
• Generative Layers, added through 2025, which keeps generated content editable as a proper layer instead of flattening it

That's a lot of tools for one bit of software. In practice, creators use maybe three of these on a regular week. The rest are there for the moment you need them. An AI product photography solution can similarly bring multiple AI-powered image capabilities into a more focused workflow.

Generative Fill: the one everyone talks about

Generative Fill is the tool that got Photoshop back in every marketer's newsfeed. You make a selection, type a prompt (or leave it blank), and Photoshop generates three variations that blend into your photo, matching lighting, shadow, and perspective as it goes.

I use it constantly for removing background clutter from event photography. Last year I had a set of conference shots for a client where every single frame had a fire exit sign glowing red behind the speaker's head. Fifteen years ago that would have meant cloning it out frame by frame, a good ninety minutes of tedious work. With Generative Fill I selected the sign in each image, typed nothing (leaving the prompt blank tells Photoshop to fill contextually), and had all twelve images cleaned in under ten minutes. That's the honest use case: not making new worlds, just quietly fixing the world you already photographed.

Here's the exact process I use on a product shot with a distracting background element:

• Open the image and select the Lasso or Object Selection tool
• Draw loosely around the object you want gone, no need for precision
• Click Generative Fill in the contextual taskbar
• Leave the prompt field empty for removal, or type a short description if you want something added instead
• Wait roughly 10 to 15 seconds for three variations to generate
• Click through the variations panel and pick the one that matches your lighting
• Flatten only once you're happy, since Generative Layers keeps it editable until then

That's the whole workflow. No prompt engineering course required.

Generative Expand and why it's better for social crops than you'd think

Generative Expand extends the edges of an image, filling in new background where none existed. This is the tool I reach for when a client sends me one perfectly composed square image and needs a vertical version for Stories and a horizontal one for a LinkedIn banner, all by end of day.

If you're already juggling formats for Instagram, it's worth pairing this with a proper look at how Instagram Story frames work, because Expand solves the image problem but framing and safe zones are still a separate issue you need to plan for.

Where Expand struggles: repeating patterns, tiled textures, and anything with straight architectural lines. Ask it to extend a brick wall and you'll often get bricks that subtly warp halfway across. Check the seams before you send anything to print.

Neural Filters, and the one I'd tell you to avoid

Neural Filters is a separate panel with filters like Skin Smoothing, Smart Portrait (which can shift someone's gaze, age, or expression), Colorize for black and white photos, and Photo Restoration for old scanned prints.

Skin Smoothing and Colorize are useful and low risk. Smart Portrait is where I'd slow down. It can change how open someone's eyes look or shift their expression toward a smile they didn't give. I've seen brands use this on team headshots and end up with photos that look subtly wrong to anyone who's met the person in real life. Small changes to a real human face carry more risk than most style guides account for, and nobody's writing that into the tutorials because it doesn't sell software.

Select Subject, Select Sky, and Sky Replacement

These are the quiet workhorses. Select Subject builds a clean mask around a person or product in one click, something that used to take fifteen minutes with the pen tool. Select Sky does the same for skies specifically. Sky Replacement then swaps a flat, grey sky for a dramatic one, adjusting the colour cast on the rest of the image to match.

Estate agents, product photographers, and travel content creators lean on this trio hardest. If you're shooting outdoors on a budget and can't wait for golden hour every time, this is the closest thing to a reshoot without leaving your desk.

Text-to-image on the canvas, and Generative Layers

Photoshop can now generate an entire image from a text prompt directly onto a blank layer, no photo needed to start from. It runs on the Firefly Image model, and Adobe lets you pick between model versions depending on which plan you're on.

This is where the honest comparison matters. If pure image generation from a prompt is your main job, standalone tools built for that purpose often beat Photoshop's version on speed and prompt understanding. I wrote a full comparison of AI image generators for business use after testing several side by side, and Photoshop's built-in generator wins on integration but not on raw output quality against the best dedicated tools. I also cancelled Canva for a stretch after testing what ChatGPT's image tool could do inside a normal workflow, which I covered in my ChatGPT Images 2.0 review. Photoshop's strength isn't generating from nothing. It's editing what you already have.

Generative Layers, rolled out through 2025, fixes the biggest early complaint: generated content used to flatten straight into your image, killing any chance to tweak it later. Now it sits as its own layer, so you can adjust opacity, mask it, or regenerate without starting the whole edit again.

What this costs

Photoshop's single-app Creative Cloud plan runs around £23.97 a month in the UK, or you get it bundled into the All Apps plan at a higher monthly rate. The generative features run on a credit system: your plan includes a set number of generative credits each month (the single-app Photoshop plan includes 500), and each Generative Fill, Expand, or Remove action uses one credit. Run out mid-month and Adobe doesn't cut you off, generation just runs slower using a lower-priority queue, or you can buy extra credit packs.

For a solo creator posting three or four times a week, 500 credits rarely runs dry. For an agency running client work across multiple brands, it goes fast, and that's worth knowing before you quote a client a flat monthly retainer that assumes unlimited AI editing.

The bit nobody puts in the marketing copy

Here's what I've learned after two years of using these tools on actual client work: Generative Fill is brilliant at removing things and mediocre at inventing them. Ask it to delete a bin from a street scene and it nails it nine times out of ten. Ask it to add a plausible-looking crowd of people in the background of an empty stadium shot and you'll spend twenty minutes regenerating before you get something that doesn't look faintly wrong around the hands and faces. The tool is at its best doing subtraction, not addition, and most of the tutorials online show you the addition tricks because they demo better on video, not because they work better in practice.

There's also a copyright question worth sitting with. Adobe trains Firefly on Adobe Stock and licensed content, which is meant to make output commercially safer than tools trained on scraped web images. That's true as far as it goes, but "safer" isn't "risk-free," and if you're producing anything for a big brand campaign or paid ad spend, it's worth having a human check the output against existing trademarks and recognisable faces before it goes live, not just trusting the model card.

None of this means skip the tools. It means use them the way I do: as a fast fix for a flawed photo, not as a replacement for taking a decent photo in the first place. If you're building out a wider content toolkit around this, it's worth browsing tools well beyond the obvious ones too, and I keep adding to a running list of useful websites most creators have never heard of that solve adjacent problems Photoshop doesn't touch. And if Instagram is where most of your edited images end up, it's worth pairing Photoshop's fixes with a proper look at the wider Instagram tools and apps that handle scheduling, captions, and analytics, because a perfectly edited image posted at the wrong time still underperforms.

For contributors covering this area, read pitching an AI article here.

Getting found early matters, so see my list of free directories for listing an AI tool.

Frequently asked questions

Do I need the full Creative Cloud plan to get Photoshop's AI tools?

No. The single-app Photoshop plan includes Generative Fill, Expand, Remove, Neural Filters, and the generative credit allowance. You only need the full All Apps plan if you want Firefly credits shared across other apps like Illustrator or Premiere Pro too.

Is Photoshop's AI good enough to replace product photography?

For small fixes like backgrounds, blemishes, and stray objects, yes. For generating a product from scratch that needs to match exact packaging, colour, and branding, no. It's an editing tool first and a generation tool a distant second.

What happens when I run out of generative credits?

Generation doesn't stop, it just slows down and moves to a lower priority queue, or you can buy additional credit packs from Adobe if you need full speed for a busy month.

Can clients tell if I've used Generative Fill on their photos?

Usually not, if you use it well and check the edges and lighting match. Content credentials metadata is attached to generated content, but it's easy to strip and most clients never check it. Disclose it anyway if it's a big enough edit to matter, it's a trust thing more than a technical one.

Sources worth reading

• Anthropic's own documentation
• the NIST AI Risk Management Framework