How to remove text from a photo

Watermark text across stock photos, date stamps burned into vacation shots by old cameras, caption overlays on screenshots, brand text on product images, graffiti text on walls in real-estate listings — text overlaid on photos is one of the most common removal targets. Magic Eraser's AI brush removes the text characters and reconstructs the underlying surface — sky, skin, fabric, wall, product surface — as if the text was never there.

Last updated May 16, 2026

Common scenarios where text needs to come off a photo — Magic Eraser

Text ends up overlaid on photos through many workflows. Stock photo previews carry watermark text across the center of the image — photographers who purchased a license for a different crop sometimes need the watermark removed from a wider view. Date stamps from older digital cameras and phone camera settings burn orange or yellow date/time text into the bottom corner of every photo, permanently marking images that are otherwise perfectly usable. Social media screenshots carry platform UI text (usernames, timestamps, engagement metrics) that the underlying photo doesn't need. Meme generators add caption text to source photos, and sometimes the clean original is lost. Product photos for resale carry the original brand's overlay text that the reseller needs to replace with their own branding. Real-estate and travel photos capture signs, graffiti, and posted notices that distract from the scene. In each case, the underlying image is what matters, and the overlaid text is an obstruction. Manual text removal in Photoshop uses the clone stamp on each character, matching the underlying surface texture stroke by stroke — a process that's tedious for large text blocks and error-prone on complex surfaces where the text crosses multiple texture regions (sky meeting a tree line, skin with tone variation, a product with surface detail). Magic Eraser's AI handles this by treating the text as an occlusion layer and reconstructing the surface beneath it from context, similar to how it handles fence removal — the regular shapes of text characters make the occlusion pattern predictable.

ステップバイステップ手順

1. 1

   Upload the photo

   Open Magic Eraser on web, iOS, or Android. Drop in the photo with the text you want to remove — the stock photo with watermark text, the vacation photo with a date stamp, the screenshot with UI overlays, the product image with brand text. JPEG, PNG, HEIC, and WebP supported.

2. 2

   Brush over the text

   Paint over each text element you want to remove. For watermark text that spans the full image, brush across the entire watermark area in one continuous stroke. For date stamps in a corner, a quick brush pass over the date string is all that's needed. For scattered text elements (multiple labels, captions at different positions), brush each one individually. Cover each character completely plus a thin margin to catch anti-aliased edges and any shadow or glow effect the text has.

3. 3

   Tap Erase and verify

   The AI removes the text and reconstructs the underlying surface. For text on simple backgrounds (sky, plain walls, solid product surfaces), the reconstruction is seamless. For text crossing complex backgrounds (text over a person's face, text spanning a detailed scene), the AI rebuilds each background region the text was covering using adjacent clean areas as reference. Check the result at full zoom — if any character remnants or edge artifacts remain, a second brush pass over those specific areas cleans them. Export at full resolution.

こんな方におすすめ

• Stock photo watermarks that obstruct the image content and need removal after license purchase for a different crop or size
• Date stamp removal from older digital camera photos where the camera burned date/time text into every shot permanently
• Screenshot cleanup where platform UI text (usernames, timestamps, metrics) clutters the underlying image content
• Product photo rebranding where the original brand's overlay text needs removal before the reseller adds their own
• Real-estate and travel photography where signs, posted notices, and graffiti text distract from the scene
• Meme and social media image cleanup where caption text needs removal to recover the clean original photo

ベストな結果を得るコツ

Text removal is generally easier than arbitrary object removal because text characters have predictable shapes with thin strokes that occlude relatively small amounts of the underlying surface. The AI reconstructs more surface than it removes, which is favorable for quality. Large block text (watermarks covering 30%+ of the image area) is harder than small text (date stamps, short labels) because more surface needs reconstruction with less adjacent reference. For large watermarks: if the watermark is semi-transparent (common for stock photo previews), the AI handles it well because the underlying surface shows through partially, giving strong reference for reconstruction. For opaque text on complex backgrounds, work in sections — brush one line of text at a time rather than the entire watermark block at once, giving the AI maximum adjacent reference for each reconstruction pass. For text with strong effects (heavy drop shadow, outer glow, emboss): brush the effect area too, not just the character strokes — the shadow or glow is part of the occlusion that needs removal. For text that repeats a background pattern (text stamped on fabric, text etched into glass): the AI often produces excellent results because the pattern provides strong reconstruction reference on both sides of each character.

よくある質問

Is it free to remove text from a photo?

Yes. Magic Eraser's free tier includes the brush tool for text removal with daily usage limits. Premium ($29.99/year) removes limits and enables high-resolution exports.

Can it remove watermarks from stock photos?

The tool removes any visible text or overlay you brush. For stock photo watermarks: semi-transparent watermarks that show the underlying image through the text remove cleanly because the AI has partial reference for the occluded surface. Opaque watermarks on complex backgrounds produce good results but larger blocks may need a section-by-section approach. Note that removing watermarks from images you haven't licensed may violate copyright — use this for images you own or have a license for.

Does it work on handwritten text and graffiti?

Yes. Handwritten text, graffiti, drawn annotations, and irregular text styles all respond to the brush tool. The AI treats any painted-over content as an occlusion to remove — it doesn't need to recognize the text as readable characters, just identify that it's overlaid content occluding the surface beneath.

What if the text crosses over a person's face?

The AI reconstructs the face beneath the text using the visible portions of the face as reference. For light text (thin strokes, small font) crossing a face, the reconstruction is typically seamless. For heavy text (large watermark, thick strokes) covering a significant portion of the face, the AI generates plausible facial continuation that may not perfectly match the original features — the result looks natural but fine details in heavily occluded areas are reconstructed rather than recovered.

Can I remove date stamps from old digital photos?

Yes. Date stamps burned into photos by older cameras are one of the cleanest removal cases because the stamp is typically 10-15 characters of small, thin text in one corner of the image with a consistent background beneath it. Brush over the date string and the AI fills with the background content. Batch-removing date stamps from a vacation album of 50-100 photos takes about 15-20 minutes working through them individually.