Advertisement 2

How to Remove Text from Image Without Removing Background in Photoshop: My Messy, Tried-and-Tested Method

July 3, 2026

·

Generative AI, Generative Fill, Photo Editing

Look, I’ve been there. You’re staring at a poster—maybe it’s a client’s event flyer, maybe it’s a vintage concert poster you’re trying to restore—and there it is. A giant block of text smack in the middle of the most beautiful background you’ve ever seen. And you think, “Okay, I’ll just erase it.” Two minutes later, you’ve got a crater where the text used to be, a weird blob of mismatched pixels, and a background that looks like it got attacked by a blind squirrel with a paintbrush.

That was me, three years ago. Until I figured this out.

I’ve tested pretty much every method under the sun for how to remove text from image without removing background in Photoshop. Content-Aware Fill? Tried it. Works great on solid colors—like, if your background is a flat blue sky, go nuts. But throw a gradient or some texture in there? Good luck. I once spent two hours trying to Content-Aware Fill a text block off a marble-textured wall. The result looked like someone had smeared toothpaste over the letters and called it a day.

The Clone Stamp? Oh man. Don’t even get me started. I had a project—a poster with this gorgeous wood grain background and some promotional text plastered right over the prettiest knot in the wood. I spent forty-five minutes sampling, Alt-clicking, painting, re-sampling, re-painting. My hand was cramping. My eyes were crossing. And you know what? I still had a visible seam where the text used to be. That’s when I realized—there had to be a better way.

Then I found Generative Fill. And honestly? It changed everything.

Here’s the thing about Generative Fill that nobody tells you.

So I’m watching this tutorial from CG Ian—you know, the one where he pulls up a poster, grabs the Magic Wand, and just poof—text gone, background perfectly intact. And I’m sitting there, coffee in hand, thinking, “No way. This is too easy. There’s gotta be a catch.”

There isn’t.

But let me tell you what I did learn the hard way. Because the first time I tried it, I just selected the text—barely, I mean I was sloppy—and hit Generate. And Photoshop gave me back this… this thing. It looked like the AI had tried to fill the space with abstract art. There were weird color blobs, jagged edges, and honestly? It looked worse than the text.

That’s when I went back and actually watched the tutorial properly. And here’s the part that matters: you have to expand your selection.

The step that saves your background every single time.

In CG Ian’s video, he selects the text with the Magic Wand—super quick, super easy. But then he goes to Select > Modify > Expand and bumps it out by 10 pixels. Ten pixels. That’s it. That’s the secret sauce.

Why does this matter? Because when you expand the selection, you’re giving Generative Fill a little bit of breathing room. You’re letting it see the background around the text—the actual texture, the actual colors, the actual lighting—so it knows what to fill in. Without that expansion, the AI is basically guessing blind. With it? It’s like giving the AI a cheat sheet.

I tested this. I did a side-by-side—one selection with no expansion, one with 10 pixels. The difference was night and day. The expanded one blended so seamlessly I literally couldn’t tell where the text had been. The unexpanded one? Looked like a bad Photoshop job from 2004.

The “Firefly” moment—and why I almost missed it.

Here’s another thing I almost screwed up. In the video, CG Ian mentions switching the AI model to “Firefly Fill & Expand 2026”. And I’m gonna be honest—the first few times I did this, I completely ignored that step. I just used whatever default model Photoshop had loaded up.

Big mistake.

The default model is fine for simple stuff. But for text removal? Especially if your background has any kind of complexity—gradients, patterns, textures—the Firefly model is just… better. It’s smarter. It understands context in a way the older models don’t. One time I was removing text from a poster with this insane sunset gradient—purples melting into oranges melting into pinks—and the default model gave me a flat, dead gray patch where the text used to be. Switched to Firefly, ran it again, and it perfectly recreated the gradient. I almost cried. Not even kidding.

My full workflow—the one that actually works.

Okay, so here’s what I do now. Every single time. No exceptions.

Open the image. Drag and drop into Photoshop. Simple.
Select the text. I use the Magic Wand—it’s fast, and as long as your text contrasts with the background, it works like a charm. If the text is on a busy background, sometimes I’ll switch to the Lasso Tool, but honestly? Magic Wand does the job 90% of the time.
Expand the selection. Select > Modify > Expand. 10 pixels. Always. This is non-negotiable for me now.
Open the Contextual Task Bar. Window > Contextual Task Bar. This is where the magic happens.
Click Generative Fill. And here’s the trick—I leave the description bar completely blank. I know some people type stuff in, but in my experience, letting Photoshop figure it out based on the surrounding pixels gives the most natural results.
Switch to Firefly Fill & Expand 2026. This is the model that actually understands what you’re trying to do.
Hit Generate. Wait about 5-10 seconds. Photoshop gives you three variations to choose from. Pick the one that looks best.

That’s it. The whole thing takes maybe two minutes. Tops.

But wait—let me tell you about the time it didn’t work.

Because it’s not always perfect. I remember this one project—a poster with this super intricate floral background. Roses, vines, the whole deal. And the text was right in the middle of a cluster of flowers. I did my usual routine—Magic Wand, expand 10 pixels, Generative Fill, Firefly model—and the result? It filled the space with… grass. Just random green grass. In the middle of a rose bush.

I was so mad. I almost threw my mouse across the room.

But then I realized—the selection was too big. I’d expanded it to 10 pixels, but because the background was so detailed, the AI didn’t have enough consistent surrounding pixels to work with. So I went back, made a tighter selection—still expanded, but only 5 pixels this time—and tried again. Perfect. The AI filled in the missing flowers and vines like they’d never been gone.

That’s the thing about this method. It’s not magic. It’s a tool. And like any tool, you have to learn its quirks. Sometimes you need to expand more, sometimes less. Sometimes you need to run it twice—once to remove the text, once to clean up any weird edges. But once you get a feel for it? It’s faster and cleaner than any other method I’ve tried.

Why I’ll never go back to the old ways.

Look, I’m not saying Content-Aware Fill or the Clone Stamp are useless. They have their place. If you’re working with a super simple background—solid color, maybe a simple gradient—Content-Aware Fill is fine. And if you’re doing hyper-detailed restoration work on a tiny area? The Clone Stamp gives you control that AI just can’t match.

But for everyday text removal? For posters, flyers, social media graphics? Generative Fill is my go-to. It’s faster, it’s cleaner, and—once you learn the expansion trick and the Firefly model switch—it’s just as reliable.

The one thing I wish someone had told me sooner.

Here’s the thing nobody mentions: always duplicate your layer first. Ctrl+J. Every single time. I cannot tell you how many times I’ve forgotten this, run Generative Fill, hated the result, and realized I’d just nuked my original background layer. Now I do it out of habit. It takes two seconds and saves you from so much heartache.

Also? Don’t be afraid to try all three variations Photoshop gives you. Sometimes the first one looks great. Sometimes the third one is the one that actually matches your background. I’ve had projects where I cycled through all three, picked the best one, and then ran Generative Fill again on a tiny leftover edge. It’s okay to be picky.

Watch someone do it—it clicks faster than you think.

I’m a visual learner. Reading about this stuff is one thing, but actually seeing someone do it—watching them select the text, expand the selection, hit Generate, and watch the text disappear like it was never there—that’s what made it click for me.

So here’s the tutorial that finally got me there. CG Ian walks through the whole process in under two minutes. No fluff, no filler. Just straight-up “here’s how you do it”. I’ve watched it probably a dozen times, and I still pick up little details I missed before.

👉 Watch the full tutorial here: https://youtu.be/PbcuJXEom2c?si=0Yzf22biIaiMEXfr&t=26