How I Edit Text That Is Already Baked Into an Image

There's a specific image problem that most tools have no good answer for: the words are already part of the picture, and the original design file is gone.

A cafe renames itself, but the sign photo is already printed on the menu. An event moves a week later, but the poster only exists as a JPG. A price goes up and the shelf label is a flat photo from two years ago. In each case, re-shooting costs more than the change is worth.

Photoshop is the obvious answer and usually the wrong one. There is no text layer to click into — the letters stopped being text the moment the file was flattened. You aren't editing text, you're editing a picture of text.

That's the problem I built Reword Image around.

Why "just put a new text box on top" doesn't work

The obvious workaround is to cover the old words with a rectangle and type new ones over the top. It demos fine and falls apart immediately:

• The background is never flat. A painted sign has brush texture, a chalkboard has dust, a label has paper grain. A solid rectangle reads as a patch from across the room.
• The lighting is baked in. Real photos have a gradient across the surface, a shadow from the frame, a falloff toward the edge. A flat fill ignores all of it.
• The lettering has depth. Engraved, embossed or weathered type casts its own small shadows. New flat text sits on the image instead of in it.

So the text has to be replaced in place: the region gets rebuilt with the material, lighting and perspective it sits on.

What that looks like

A wooden hanging sign, renamed. The grain, the chain shadow and the slight perspective of the photo all have to survive:

An event poster with the date moved:

And the harder case — removing lettering entirely and rebuilding whatever was behind it:

What works, and what doesn't

Input quality dominates the result, so it's worth being specific instead of listing features.

Works well

• Large, clear lettering
• Simple to mid-complexity backgrounds
• Short replacements — a word, a name, a date, a price
• Straight-on or mildly angled shots

Needs another attempt

• Very small text (roughly under 20px in the source image)
• Curved, script or heavily decorative lettering
• Long replacements — the new text has to fit the space the old text occupied
• Lettering on a busy, high-frequency background

Not what this is

• Not a design tool. It won't hand back a PSD or recover the original font file.
• Not a general object-removal inpainter — there's a narrower eraser for lettering.
• Not a one-click fix for everything. Some images take two or three attempts.

I'd rather state that up front than let people find out after paying.

The workflow

Deliberately narrow, because a wider one would be a worse one:

1. Upload a JPG, PNG or WebP
2. Type the text that's currently in the image
3. Type the replacement
4. Generate, compare against the original, download

Step 2 surprises people. You can't drop a freely positioned new text layer, because the whole point is replacing what is already there. The trade-off pays off: the result keeps the original typography, perspective and material instead of looking pasted on.

Around that core there are three focused tools and one eraser:

• Replace text in an image — the general case
• Change a date in an image — dates, deadlines, schedules
• Edit numbers and prices — prices, quantities, opening hours
• Remove text from an image — erase lettering and rebuild the background behind it

One credit covers one complete edit, not one text box: changing a date, a price and a headline in the same image is a single credit.

Pricing, and two deliberate choices

The first edit is free at 1K output — sign in with Google to claim it, no payment method needed. After that it's one-time credit packs: $4.99 for 20 edits, $19.90 for 100, $49.90 for 300, each returning 2K output. No subscription.

Two things I decided early:

1. Credits don't expire. A subscription for a tool you use twice a quarter is a bad deal.
2. A technically failed edit returns the credit automatically. If the pipeline errors, that isn't the user's problem to pay for. Plenty of pay-per-image tools keep the credit; I didn't want to run one of those.

Try it

rewordimage.com — English and Spanish.

If you have a flattened image with the wrong words in it, that's exactly the case it's built for. And if it fails on your image, I'd genuinely like to see what the input looked like — the failure cases are the ones I need most.