Edit text in images

Rewrite the words inside an image with AI while keeping the original lettering style and layout.

What edit text does

Edit Text rewrites the words inside an image — a poster headline, a product label, a banner caption — while keeping the original lettering style, color, and layout. Lovision detects the text in the image, you rewrite any of the detected blocks, and a new image is generated where only those words change. Pixels outside the edited text are taken from the original image, so the rest of the picture is preserved.

Unlike Extract text, which converts image text into editable text layers, Edit Text keeps the result as a single image with the new wording drawn in place.

Editing text in an image

Select an image layer on the canvas.
In the floating inline toolbar, click Edit text.
Wait while Lovision finds the text in the image. Detected text blocks are highlighted on the image, and a panel beside it lists each block as an editable field.
Rewrite the blocks you want to change. Anything you leave untouched is kept as-is — clearing a field does not delete the text, it just doesn't count as an edit.
Click Run. The panel closes and a loading placeholder appears beside the source image.
When generation completes, the edited image lands in the placeholder as a new image layer. The original image is untouched.

What to expect

While the session is active, the floating toolbar shows how many text blocks were found and how many you have edited, plus a Cancel button that exits the session without running.
Up to 10 blocks can be edited in a single run. To change more, run the current edits first and start a new session on the result.
Restore original text reverts a block to its detected wording; Undo last change steps back your latest edit inside the session.
The result is a new layer beside the source, so undo (⌘Z) removes it in one step.
If the model reframes the image instead of only changing the text, the result is discarded and you can try again. If it changed more than the edited words, everything outside those words is kept from the original image and a warning asks you to check the result.

Limitations

Viewers in read-only mode cannot start an Edit Text session.
Tiled image fills, split-layer images, and unsaved local sources are not supported.
You can rewrite detected blocks but not add new text blocks or draw your own regions.
Rotated or curved text is detected as an upright rectangle, so its highlight may look larger than the letters.

Tips

Keep replacements close to the original length — much longer text has to fit into the same space.
Images with crisp, readable text detect best; if the text sits in one part of the image, crop to it first.
To turn image text into real, editable text layers instead of rewriting it in place, use Extract text.