How to Edit Text in an Image With AI: Steps, Models and Tips (2026)

To edit text in an image with AI, upload the picture to an AI image editor, tell it the exact words to find and the exact words to put in their place, and let a text-capable edit model re-letter only that spot in the same font, colour and position. In Kenerate that is Edit Studio → Edit Text in Image → Replace: up to five old → new pairs per run, a Translate mode for 20 languages and a Remove mode, with GPT Image 2 as the default model. Check every letter afterwards, because small or curved text is still where AI edits slip.

Updated as of 31 October 2026. Model lists and credit costs are taken from Kenerate's live app configuration on that date.

What does "editing text in an image" actually mean?

A photo, screenshot or flattened poster has no text layer. The words are just pixels, so a normal photo editor can't click on them and retype. Until recently you had three options: rebuild the design from scratch, clone the old letters away and type new ones on top, or hand it to a designer.

AI editing adds a fourth option. An image-edit model looks at the whole picture, works out which pixels are the letters you named, and redraws those letters as different words while trying to keep the typeface, colour, lighting and perspective the same. Everything else in the frame should stay put.

People use it for jobs like these:

• Fixing a typo on a poster, flyer or AI-generated image before it goes out.
• Updating a detail: a price on a menu board, a date on an event banner, opening hours on a shop sign.
• Swapping a name on a mock-up, a product label concept or a sign in a scene.
• Translating a banner or infographic into another language without redesigning it.
• Removing text entirely, such as captions burned into a photo or placeholder lorem ipsum.

Which method should you use?

There is no single best method. The right one depends on how much text you're changing and how exact the font has to be.

A quick rule: if you can describe the change as "replace X with Y", start with an AI find-and-replace. If the font has to be pixel-exact for a brand, remove the text with AI and set the new words yourself.

How do you edit text in an image with AI, step by step?

Here is the workflow in Kenerate's Edit Studio. The same logic applies to most AI editors.

1. Open the editor and upload the image. Go to Kenerate's image editor and drop in your image (the upload limit is 50 MB). Use the largest, sharpest version you have; blurry letters give the model less to copy.
2. Pick the tool. Choose Edit Text in Image. You'll see three modes: Replace, Translate and Remove.
3. Type the old text exactly as it appears. In Replace mode, write the original words in the first box with the same spelling, capitals and punctuation. The app's own hint says it plainly: type the old text exactly as it appears.
4. Type the new text beside it. Each pair can hold up to 120 characters. Add another pair with Another word if you need to change more than one thing; you can have up to five pairs in one run.
5. Choose a model. GPT Image 2 is the default. The section below explains when to switch.
6. Choose how many takes. One, two or four variations. Two or four costs more but gives you options when the first letter spacing isn't quite right.
7. Run it and zoom in. Check each letter at 100% or more, especially numbers, accents and the spacing either side of the new word.
8. Re-run or fix locally. If one word is wrong, run that single pair again instead of all five. If a small patch around the text looks off, use the region tools (Change Region or Magic Eraser) on that spot only.

Behind the scenes, Replace mode sends the model a tightly scoped instruction: edit only the text, replace each pair, spell it exactly as given, keep the same font, size, weight, colour, effects, alignment and position, and change nothing else. That is the same structure you should use if you're writing your own prompt in another tool.

What prompt works best if you're writing it yourself?

If you use a general image model with a reference image (for example GPT Image 2.5 or Nano Banana 2.1 in Kenerate's AI image editor), write the instruction like this:

Edit only the text in this image. Replace the text "SUMMER SALE" with "WINTER SALE". Keep exactly the same font, size, weight, colour, effects, alignment and position, spelled exactly as given. Do not change anything else in the image.

Three things make it work:

• Quote both strings. Quotation marks tell the model these are literal letters, not a description.
• Name what must stay. Font, colour, size and position are the properties models drift on most.
• Close with "change nothing else". Without it, some models "improve" faces, backgrounds or lighting.

Which AI model is best for editing text in a picture?

As of 31 October 2026, Edit Text in Image in Kenerate offers five models. Costs are per take at default settings.

Start with the default. If the spelling is right but the surrounding design moved, try Nano Banana 2. If you're running many simple edits (the same price change across ten menu photos), test Qwen Image 3.0 first because it costs under half as much.

For brand-new images that need perfect text from the start, rather than editing an existing one, see our comparison of the best AI image generators for text.

Can AI translate the text in an image?

Yes. Translate mode re-letters every visible word into one language you choose and asks the model to keep the same typography style, weight, colours, sizes, effects, layout and positions. Kenerate's list has 20 languages: Hindi, English, Spanish, French, German, Portuguese, Arabic, Japanese, Korean, Chinese (Simplified), Russian, Italian, Turkish, Indonesian, Bengali, Tamil, Telugu, Marathi, Gujarati and Urdu.

Some practical notes:

• Translations change length. German and Russian often run longer than English, so the model may shrink the type to fit. Check that it's still readable.
• Scripts change shape. A bold English headline can't keep the same letterforms in Tamil or Arabic; the model picks a style that feels similar.
• Have a native speaker check it before printing. Treat the output as a strong draft, not a certified translation.

How do you remove text from an image instead?

Choose Remove mode in the same tool, or use the dedicated remove text from image page. Remove mode asks the model to clear all visible text, captions and lettering and rebuild the background behind it.

Removal is also the most reliable route for brand-critical work. Remove the old words with AI, then add your own text layer with the exact font file in Kenerate's text tool, Canva, Figma or Photoshop. You get the AI's clean background and a perfect typeface.

If you only want to remove one caption and keep the other text, use the region tools instead: mark just that area so the model leaves the rest of the lettering alone.

Why does AI sometimes misspell or warp the new text?

Image models draw letters as shapes, not as characters typed from a keyboard. Newer models are far better at this than models from two years ago, but they still slip in predictable places. We cover the technical reasons in why AI image generators misspell text. The short version for editing:

• Small text (fine print, labels on a product in the background) has too few pixels to copy cleanly.
• Curved, angled or textured surfaces (bottles, fabric, neon tubes, embossed metal) make perspective hard to keep.
• Long strings raise the odds of a doubled or dropped letter. Split long copy into shorter pairs.
• Unusual scripts and stylised display fonts give the model fewer examples to imitate.
• Repeated edits on the same file pile up small artefacts. getimg.ai's own guide notes that extensive edits, typically beyond six iterations, can introduce minor artefacts, and Descript's help centre recommends three or fewer AI edits per original image. Go back to the original file for each new attempt where you can.

Traditional tools have the same problem. In an Adobe Community thread, a Photoshop user reported that Generative Fill returned garbled words when asked to replace one word in an image; the community expert's advice was to type the replacement with the Type tool instead.

How do you match the original font?

If you want the AI to do it, give it the best possible source: a sharp, straight-on photo, the old text quoted exactly, and the instruction to keep the same font, size, weight and colour.

If you want to set the type yourself:

1. Remove the old text with AI so the background is clean.
2. Identify the font. In Photoshop, draw a Rectangular Marquee around a clear sample of the text and choose Type → Match Font. Adobe's tutorial walks through it, and suggestions come from your installed fonts plus Adobe Fonts. Community experts note Match Font works for Latin and Japanese characters only, and that straightening the perspective first helps.
3. Type the new words on a separate layer and copy any effects (shadow, stroke, bevel, glow).
4. Blend it in. Match the grain, slight blur and lighting of the photo so the new text doesn't look pasted on.

Is it okay to edit text in any image?

Editing your own posters, product shots, menus, social graphics and AI images is everyday design work. Don't use AI text editing to alter documents, receipts, IDs, certificates, contracts, invoices or other people's screenshots so they say something they didn't. That can be fraud or defamation. When you edit a photo that will be published as real (news, reviews, evidence), say it was edited.

What are the best tips for clean results?

• Use the original, highest-resolution file, not a screenshot of a screenshot.
• Crop to the area that matters if the text is tiny, edit, then place it back.
• Keep each pair short and change one idea per pair.
• Match capitals exactly: if the sign is all caps, type the new word in all caps.
• Run two takes on anything that will be printed; pick the cleaner one.
• For a quick, free crop, resize or filter before or after, use the free AI photo editor online.
• For the fastest route straight into the tool, start from the change text in image page.

The bottom line

AI has made "change the words in this picture" a one-minute job instead of a redesign. Type the old text exactly, type the new text, keep the instruction narrow, and zoom in before you ship. Use GPT Image 2 for spelling-critical edits, Nano Banana 2 when the layout must not move, the cheaper models for drafts, and the remove-then-retype route when a brand font has to be exact.

Sources

• getimg.ai, "Guide to using Text Editing in Image Editor", accessed 31 October 2026.
• Descript Help Center, "Edit images with AI", accessed 31 October 2026.
• Adobe Community, "Using Generative Fill AI to replace word in an image (same font, etc.)", thread from December 2023, answer September 2024.
• Adobe, "Match fonts in an image" (Photoshop tutorial).
• Kenerate app configuration (Edit Studio text-edit models, credits, modes and languages), as of 31 October 2026.