Remove text from an image

Mark the text you want gone and the model rebuilds the surface it was printed on. This is inpainting, so instead of painting a block over the words, it regenerates the wall, sky, fabric, or paper that the letters were covering. The result reads as if the text was never added, because the fill continues the surrounding scene rather than hiding the words behind a patch.

Yes. It handles both text laid over a photo, like a caption or watermark, and text that is part of the scene, like a shop sign, a poster, or writing on a wall. For scene text the model rebuilds the sign face or surface, so the object stays but the words come off.

Yes. Mark the full band of text, a caption strip along the bottom, a lower-third title, or a burned-in subtitle, and clear it in one pass. Because the model fills from what surrounds the strip, a busy background may want a rerun to keep the fill clean.

No, that is the point of rebuilding rather than covering. Older tools drop a soft blur or a solid block over the words, which is obvious. Here the region is regenerated to match the texture and lighting around it, so there is no fuzzy rectangle left behind. If a patch does come back soft, rerunning usually fixes it.

Yes. Mark only the words you want gone and leave the rest untouched. You can clear a phone number off a poster while keeping the headline, or drop one line of a caption and keep another, working line by line if the layout is tight.

Captions and subtitles, watermark text, dates and timestamps, logos with wordmarks, signage, labels, price tags, and handwriting. If you can point to the letters, you can ask for them to be removed and the surface behind them rebuilt.

No. It runs on Morphic in the browser. Upload the image, mark the text, generate the clean version, and export it from the same canvas.