The safest way to remove text from a picture is to edit only the pixels occupied by the words. An AI inpainting tool uses the visible color, texture, and structure around your selection to build a replacement. It does not uncover a hidden original.
That distinction matters. A short caption on sky, wall, or another simple surface is often a manageable local edit. Text crossing a face, product edge, or detailed pattern asks the model to invent information it cannot see. Use this workflow only on photos you own or have permission to modify.
What happened in our real text-removal test
We added two lines—“SUMMER SALE” and “30% OFF”—to a controlled photo from this project, uploaded the 1200 × 900 WebP to the live remove text from image tool, and marked the lines with two passes of a 32 px brush. The no-login workflow auto-selected LaMa Eraser and estimated two credits. One generation returned a 1200 × 900 PNG.
- Text: every visible character and outline was removed.
- Subject: the cat, crop, color, and framing stayed unchanged.
- Surface: the plain wall was continued without a hard seam.
- Honest limit: at 100% zoom, the former text area is slightly softer than the untouched wall.
We also tried masking text inside a dark rectangular badge. That attempt left letter fragments and a visible smear, so it failed our quality check and is not used as the comparison above. The lesson is practical: text over an even background is easier; text combined with a graphic shape may need the whole authorized badge removed, a smaller second pass, or the original layered design file.
How to remove text from a photo in four steps
- Start with a file you may edit. Keep the untouched original, then upload the largest clean JPG, PNG, or WebP available.
- Brush over the complete caption. Include every letter, outline, glow, and shadow without covering nearby details that should stay.
- Generate one focused edit. For several separate captions, remove and review one group at a time instead of masking half the photo.
- Compare before downloading. Inspect the result at normal size and 100% zoom for fragments, soft patches, broken lines, or repeated texture.
TRY THE SAME WORKFLOW
Brush over the caption, then inspect the repaired surface.
Remove text from your imageHow to mask text without damaging the photo
Cover the anti-aliased edge
Letter edges contain partially transparent pixels. If the mask stops inside the visible glyph, a pale outline can remain after generation. Paint just beyond the edge, including any drop shadow, but keep the margin narrow.
Use one line or short group at a time
A compact selection gives the model more untouched context. If a subtitle spans the bottom of a photo, split it where the background changes—for example, sky on one side and a jacket on the other—then review each result before continuing.
Protect important structure
Do not paint over a face, logo you need to keep, product contour, horizon, railing, or other straight edge just to make the mask faster. When text crosses one of those details, use a smaller brush and expect that the exact hidden structure cannot be recovered from the photo alone.
When text removal gets difficult
Text over faces or unique details
The model can create a plausible face or detail, but not the exact content the camera captured before the words were added. If identity or factual accuracy matters, find the original photo instead of relying on generated pixels.
Text over patterns and gradients
Brick, fabric, wood grain, and smooth gradients make subtle mistakes visible. Work in smaller sections and follow each line or repeated unit through the edited area. A clean-looking thumbnail can still hide repetition or banding at full size.
Text inside a badge, banner, or sticker
Decide whether you want to remove only the letters or the entire graphic. Our first test showed that narrow masks over letters inside a solid badge can produce smearing. If you own the creative, removing the complete badge can give the model a simpler background-reconstruction task.
The 30-second result check
- No letter fragments, outlines, glow, or shadow remain.
- Nearby faces, products, logos, and edges are unchanged.
- Gradients and repeated textures continue through the former text area.
- The repaired patch does not look noticeably softer at 100% zoom.
- The result is suitable for the final crop and display size.
If the result fails, return to the original rather than repeatedly editing a damaged output. Change the mask first: widen it to capture leftover edges, or tighten it to protect nearby structure. For broader removal advice, see our unwanted-object removal guide.
Remove text from a photo FAQ
Can I remove text from a photo for free?
You can open InpaintingAI and try the remove-text workflow without signing in. The editor shows the current free allowance before you generate an edit.
Can AI recover the exact pixels hidden behind text?
No. An inpainting model generates a plausible replacement from visible nearby pixels. It cannot reveal original detail that the text completely covered.
How much of the text should I brush over?
Cover every character plus its outline, shadow, glow, and anti-aliased edge. Keep the mask focused so nearby faces, product edges, and straight lines remain outside it.
Is removing text from a picture different from removing text from an image?
No. Photo, picture, and image describe the same pixel-editing task here. The important distinction is whether the words are burned into the pixels or still exist as an editable text layer.
Can I use this method to remove a watermark?
Only remove a watermark or ownership mark when you own the image or have permission to modify it. A watermark often crosses detailed content, so it can also be more difficult than a caption on a plain surface.