Text removal is not a magic eraser for every overlay. It is for lettering: subtitles, captions, date stamps, prices, schedule grids, and “PROOF” / “SALE” lines that sit on top of a photo you own. Logo-shaped badges belong on watermark removal. Emoji and stickers belong on emoji removal. Running the wrong model is why a pass leaves glyph ghosts or smears a brand mark. Auto text removal is the first pass when the type is readable as letters. Manual brush is for leftovers after you zoom to 100%.
Backgroundcanvas runs in the browser. Upload JPG, PNG, WebP, or AVIF up to 10MB, long edge 6000×6000 pixels. A typical 2000–3500 px JPEG under 4MB usually returns in under 60 seconds. After the job, drag the compare slider at 100% on the former letters and on a clean patch next to them. Matching grain is the pass criterion — not a blurry rectangle where the caption used to be.
Subtitles, captions, and social handles
Hard-burned subtitles sit on a relatively flat band at the bottom of a still. Auto usually clears the glyphs; inspect the band at 100% because reconstruction often leaves a slightly cleaner strip than the original compression. If you still need the dialogue, keep a copy of the original file — this tool does not extract text, it paints it out. Social handles and “follow us” lines in a corner behave like captions. Paint only the type. Leave faces, products, and logos you intend to keep unpainted. If a handle is drawn as a sticker with a white pill behind it, route that one to emoji removal after the letters are gone.
Date stamps, prices, and schedule grids
Camera-roll dates, POS prices, and gym or classroom grids are dense type. Brush the whole block, not one digit at a time, or you will get a checker of leftover stems. Zoom the former times. You want the photo or brand color behind the grid, not a soft smudge. Stacked stamps (“cancelled” over an older time) need the whole stack painted. If the grid covers most of a face, stop and use a layered source file. Reconstruction copies nearby pixels; it cannot invent an eye that was under the type.
Type you must keep
Book covers, posters, and course art often mix a title you want with a promo line you do not. Do not run full-image auto on those frames. Use Manual and paint only the dated line. If auto already ate a letter of the title, you need the designer file, not a second pass. Legal line: only edit images you created, commissioned, or were hired to retouch. Clearing your own expired promo is ops. Clearing someone else’s copyrighted type is not.
Limits, timing, and when to stop
Hard limits you can reproduce: 10MB; 6000×6000 px; JPG/PNG/WebP/AVIF only. A 2400 px JPEG with a bottom caption typically finishes in under a minute. A dense A2 poster scan near 10MB can take longer and often needs a second brush pass on leftover glyphs. Skip the tool on screenshots of screenshots with heavy blocking, on type that is the subject (a storefront sign you need sharp), and on files you do not own. Then download HD. If you still need a wider canvas, finish text removal first — outpainting a captioned frame copies the letters into the new pixels.
テキストと背景
Backgroundcanvas で画像からテキストを除去するには？
テキスト除去ツールを開き画像をアップロードすると、AI が字幕・キャプション・透かし文字を消去します。結果を比較して HD 画像をダウンロード。
Full-image auto can clear every readable line. If the frame mixes a title you need with a promo you do not, use Manual and paint only the promo. If auto already damaged the title, go back to the layered design file.
How long does text removal take?
A typical photo under 4MB and about 2000–3500 pixels on the long edge usually finishes in under 60 seconds. Dense poster scans near the 10MB or 6000 px cap can take longer and often need a manual pass on leftover glyphs.
