Edit Text in an Image
Type what the picture says now and what it should say instead. The AI rewrites it in the same font, colour and perspective, and leaves the rest of the image alone.
Leave this empty to remove the text instead of replacing it.
Result
In short
To change the words inside a photo: upload it, type the text exactly as it appears, type the replacement, and press Edit text. Each edit costs 3credits and takes 10–30 seconds. If you only want the words gone, the free Remove Text tool does that better and costs nothing.
When this is the right tool
A price or a date that changed
The menu board says £8.50 and it is now £9. The poster says Thursday and the event moved. Reshooting the photo is not an option; retyping two characters is.
A typo you only noticed after export
The graphic is signed off, the source file is on someone else's laptop, and there is one wrong letter in a headline. This fixes the flattened image directly.
Swapping a name or a label
The same mockup with a different brand on the packaging, a different room number on the door, a different word on the badge.
Localising a single word
One sign in the background of an otherwise reusable photo. Change it rather than rebuilding the whole shot.
Editing text is a different job from erasing it
Editing text vs erasing it
Erasing is the easier problem: the tool rebuilds the surface the lettering sat on and the words are simply gone. That is the free Remove Text tool, and if you want the words gone rather than changed, use it — it is quicker, cheaper, and better at exactly that job.
Why editing costs more
Replacing text means rendering new glyphs that match the original's font, weight, colour, spacing, perspective and any curvature or texture, then setting them into the scene convincingly. That needs a much larger model than an inpainting pass, and it is billed per call, which is why this tool is credit-priced while erasing stays free.
What it changes
The instruction sent to the model is deliberately narrow: replace this specific string, match the original lettering, leave everything else alone. It is still a generative edit, so the whole image is re-rendered rather than patched — expect very close, not pixel-identical outside the text.
Five things that decide the result
- 1
Type the text exactly as it appears, including capitalisation and punctuation. The model finds it by matching, so ALL CAPS matters.
- 2
One string at a time. If a poster has three things to change, run three passes — a single instruction with three swaps gets muddled.
- 3
Keep the replacement a similar length. Fitting a much longer phrase into the same space forces the model to shrink or crowd the lettering.
- 4
Short, high-contrast text on a flat surface works best. Small text, heavy texture, or a strongly curved surface are the hard cases.
- 5
Leave the second field empty to delete the text instead — though the free Remove Text tool usually does a cleaner job of that.
Questions about editing text in images
Other ways to deal with text
Powered by apimodels.app — a unified AI image generation API