사진에서 텍스트 제거하기

This URL is the editor, not a brochure that sends you to another app. You drop a picture above, choose auto or brush, and download a clean frame without leaving the page. Searchers who type “remove text from photo” usually want that loop to finish in the same tab — upload, erase, compare, save.

Lettering shows up in places Photoshop used to own: a shop neon, a sale sticker in a window, a timestamp in the corner, a caption burned into a screenshot, a watermark you no longer need. Cloning by hand is slow when the background is brick, foliage, or glass. Inpainting estimates what sat behind the glyphs and paints it back so the edit reads as a photograph, not a smear.

Auto mode is for dense signage you want gone in one pass. Brush mode is for a single headline on a poster, a name on a jersey, or a date stamp you must not touch the rest of the frame for. Both paths spend credits shown on the button before you run. New accounts get signup credits so you can test on a sample or your own file first.

Uploads are processed to complete the job and are not kept as a permanent library. Files are removed from our systems within a day, and we do not train models on your pictures. Download the result if you need it later — the source of truth stays on your disk, not in this tab.

Auto mode scans the whole frame. It is the right default when a storefront, billboard, or meme is covered in type and you want a clean scene. It can also catch small stamps you missed. If you only need one word gone and must keep a logo elsewhere, skip auto and paint the target.

Brush mode is a mask. White (painted) pixels tell the model where to inpaint; everything else should stay. Keep the stroke slightly larger than the letters so serifs and drop shadows are included. If you paint a face or a product label by accident, switch to eraser before you generate — credits are spent on the mask you submit, not on a second thought after download.

Credit cost differs by path: auto is the heavier full-image pass; brush is cheaper because the job is local. The footer of the editor shows the amount before you click. Sign in is required so usage can be metered; signup credits cover first tries.