Remove Text
Erases one existing text block and regenerates the background without adding new wording.
AI image text removal
Remove text from an image without Photoshop. This image text remover erases one selected block of words, such as a caption, a label or a date stamp, and rebuilds the background behind it. No replacement text is added. Upload a JPG, PNG or WebP up to 12 MB, pick the text, review the result and download a flattened PNG.
This workflow is intended for images you own or have permission to modify. It does not remove video subtitles, edit document records, or provide a general-purpose object removal brush.
Choose a clear image containing the complete caption, date, label, headline, or short text block you want to erase.
Real before and after
Compare the original PRECISION image with the flattened result produced by the current Remove Text workflow.


This example was generated with the current Image Text Editor workflow in Remove Text mode using the original uploaded image. One detected text block was removed and the selected area was reconstructed using the surrounding image as the visual reference. The result is a flattened image, not an editable removal mask or text layer. Results vary with image quality, text style, background complexity, and model output.
Each pair below is a real result from the current Remove Text workflow, generated from the original uploads and published without retouching. One detected block was removed per image. Results vary with image quality, text style, background complexity, and model output.
Before
After
Before
After
Before
After
Before
AfterPixels, not a live text layer
Text inside a finished JPG, PNG, or WebP is part of the image pixels. Removing it is not the same as deleting a live text layer. The selected letters, outlines, and shadows must be regenerated, and the missing area is filled using the surrounding image as a visual reference. This reconstruction process is often called inpainting.
There is no separate layer to delete. A headline on a travel photo, an orange date imprint on an old print, and hand-painted writing on a shop window are all part of the same pixel grid, so removing words from a picture means rebuilding whatever sat behind the letters. The model works from the surrounding pixels rather than the original scene, which is why a plain wall is usually easier to rebuild than a face, a reflection, or a repeating pattern.
Choose the correct workflow
Erases one existing text block and regenerates the background without adding new wording.
Removes the old wording and generates a new text block in the selected area. Replace text in an image.
Places a shape, blur, crop, or another element over the wording without reconstructing what was behind it.
The text remover works on finished images where the words are part of the pixels: a photo with an old caption, a screenshot with a label you no longer need, or a social graphic with outdated promo copy. Select the text, remove it, and the area is filled to match its surroundings.
If you want new words in the same spot, replace text in the image instead. To change one word or a headline while keeping the layout, start from the image text editor. After the old words are gone, you can add text to the photo for free.
Authorized cleanup tasks
Remove an old caption, headline, slogan, or short promotional message from an image you own.
Erase a date stamp or timestamp from your own photo when it is no longer needed.
Remove placeholder wording from an authorized template image before preparing a new version.
Clear an expired offer, event message, campaign label, or short promotional line from your own graphic.
Remove a visible label, annotation, note, or short callout from an image you are authorized to edit.
Clean one embedded caption or campaign phrase from a social media image before reusing the visual.
Background reconstruction
Six-step text removal
The same steps apply whether you are removing a printed date from a photo, a few words of writing on a shop window, or one line of letters from a product image. Removal targets a detected block, so what changes between cases is the surface behind the text, not the workflow.
Keep an untouched copy of the JPG, PNG, or WebP. Every removal returns a flattened PNG, so the original file is the only way back to the words you erase.
Drop the file into the editor (up to 12 MB). The scan lists readable text blocks, including captions, headlines, date stamps, and painted signs.
Choose the block that holds the letters, words, or date you want gone, and check that the highlighted box covers the full text and nothing you want to keep.
Include outlines, shadows, glows, and any underline that belongs to the text. A tight box around the letters alone can leave a visible edge behind.
Confirm the removal. Nothing is typed over the area; the model rebuilds the background from the pixels around the selection. Use Standard 1K for a quick check or HD 2K for a final asset.
Open the before-and-after view at 100% and check edges, texture, lighting, and nearby text. If a patch shows, start again from the original file, then download the flattened PNG.
Matching new lettering to the words you keep is a separate task; edit text in an image with the same font style uses the original lettering as a visual reference.
Select the complete visible effect
Select the entire word, caption, date, label, or headline rather than only the center of each character.
Review the flattened result
Look for leftover strokes, punctuation, outlines, shadows, glows, or partially erased characters.
Check for rectangular fills, blurry patches, repeated texture, broken gradients, or abrupt color changes.
Inspect faces, hands, products, signs, logos, edges, and other meaningful objects near the removed text.
Review brick, fabric, wood, tiles, leaves, grids, and other repeating patterns.
Confirm that highlights, reflections, light direction, and natural shadows remain consistent.
Check that nearby captions, labels, numbers, and unrelated wording were not changed.
Inspect the result enlarged to find small artifacts around the removed area.
Preview the image at its intended mobile, social, marketplace, web, or print size.
If the removed text sat over a pattern or a gradient, the review notes for replacing text without changing the background cover the same checks: gradients, repeated texture, and edges that no longer follow the surface.
Flattened output
The downloaded PNG does not contain a reversible removal mask, editable background layer, live text object, or controls for restoring the removed wording afterward.
Keep the original image. Use the before-and-after comparison to review the result before downloading.
Limits and authorization
Results are most consistent when the text is sharp, readable, and placed over a simple or moderately detailed background. Very small letters, curved text, reflective surfaces, dense patterns, strong perspective, and large selected regions can be less predictable. The current workflow focuses on one selected text block per generation.
Only upload and edit images you own or have permission to modify. Do not use this tool to remove watermarks, copyright notices, creator credits, signatures, official seals, verification marks, or ownership labels without authorization.
Do not use it to alter identity documents, receipts, invoices, certificates, tickets, contracts, financial records, screenshots used as evidence, or other materials intended to prove facts or authorization.
Image text removal questions
You can try it for free. Eligible new accounts get one free credit after Google sign-in, which covers one Standard 1K removal. After that, credit packs start at $4.99 for 12 credits.
Upload the picture here, select the detected text block, and check that the highlight covers the whole text, including any shadow or outline. Then run the removal. You do not need layers, masks or the original design file.
Yes. Upload the finished image, select one detected text block, confirm the removal, and generate a new flattened image with the selected area rebuilt.
No. Remove mode deletes the selected wording and reconstructs the background without adding new text.
Yes. Removing text leaves the selected area without wording. Replacing text removes the old wording and generates new text in the same area.
Yes, when the word you want gone is returned as its own block. The scan can detect words set apart on a line as separate blocks, so the highlight may cover one word instead of the whole line. Adjacent lines are also grouped into a paragraph candidate, so one line of a paragraph can be removed on its own or the whole paragraph in a single pass. Check the highlighted box before you confirm: if it covers the whole line, the whole line is the target.
Exact preservation cannot be guaranteed. The selected area is regenerated using the surrounding image as a visual reference, so you should inspect textures, gradients, lighting, edges, and nearby objects.
Nothing is pasted on top. The selected area is regenerated by the image model from the surrounding pixels, the process described above as inpainting. Flat walls, open sky, and smooth gradients are usually rebuilt more predictably than dense patterns, faces, glass reflections, or text over a person. Exact preservation is not guaranteed, so compare the result at full size before publishing.
It may be possible, but results can vary when text overlaps a face, hand, product, logo, or another meaningful subject. Review the generated image before using it.
The current workflow handles one selected text block per generation. Remove and review separate areas one at a time.
The current workflow relies on detected text blocks. Try a clearer, higher-resolution source image when the required wording is not detected.
Yes. In testing, a small, faint date imprint was detected and removed like any other block, so an old-style orange date stamp is a reasonable candidate. Detection is still the gate: if a stamp is not returned as a block, try a sharper scan of the same photo, as described above.
Do not remove watermarks, copyright notices, creator credits, or ownership labels unless you own the image and have clear authorization to do so.
No. The downloaded result is a flattened image. Keep the original file if you may need the removed wording later.
No. The current workflow edits static JPG, PNG, and WebP images rather than video frames or subtitles.
The editor accepts JPG, PNG, and WebP files up to 12 MB, whether the text sits on a photo, a scanned print, or a screenshot. The result is downloaded as a flattened PNG, so keep the original file if you may need the removed words again.
Eligible new accounts receive 1 free credit, available for one Standard 1K result, plus capacity for 3 image scans. Paid one-time credit packs start at $4.99 for 12 credits; monthly and yearly subscriptions are also available.
Original uploads are stored privately for up to 24 hours. Completed flattened results and related private diagnostics are kept for up to 7 days unless you delete them sooner. Your uploaded image and relevant text instructions are sent to the providers needed to perform the edit. Read the Privacy Policy.
Removal works on JPG, JPEG, PNG, and WebP files. To put new wording in place of the text you erased, use the JPG text editor or the PNG text editor; for interface text in a screenshot, start with the screenshot text editor.
Start with the finished image
Upload a finished image, select one caption, date, label, headline, or short text block, remove it, and review the rebuilt background before downloading.
Upload an Image to Remove Text