無料 · 画像
画像からテキスト（OCR）.
写真・スクリーンショット・スキャン画像から、日本語やアラビア語を含む 11 言語の OCR でテキストをローカル抽出。確認・修正してコピー、または TXT や Word に書き出せます。
JPG, PNG, WebP, BMP or GIF (first frame) · up to 10 images, 15 MiB each, 60 MiB per run, 25 megapixels per image. Images stay in your browser; the OCR engine and selected language models load from this site (about 2–20 MB per model, cached after first use).
Printed text, screenshots and clear photos work best. Handwriting, decorative fonts, low resolution, blur, perspective and complex tables are unreliable. Recognition does not rebuild layout, tables or formatting. Always review the text before relying on it.
使い方: 画像からテキスト（OCR）
データを入力するかファイルを選び、設定を調整して結果を生成します。コピーやダウンロードの前に結果を確認してください。
制限と対応形式
入力内容はお使いの端末で処理されます。
英語の技術説明
JPG, PNG, WebP, BMP or GIF first frame; up to 10 images, 15 MiB each, 60 MiB and 500,000 characters per run, 25 megapixels per image. Images are reduced to about 12 megapixels for recognition. Printed text only; handwriting, decorative fonts, blur, perspective and complex tables are unreliable. No layout, table or formatting reconstruction.
Recognition uses Tesseract LSTM models (tessdata_best) served from this site: English, French, Arabic, Spanish, German, Portuguese, Italian, Dutch, Hindi, simplified Chinese and Japanese, up to three per run. Images are decoded with their EXIF orientation, optionally rotated, enlarged when text is small and cleaned up: automatic mode converts to grayscale, stretches contrast and inverts light text on dark backgrounds. Layout modes cover automatic page segmentation, a single column, a single block or a single line; automatic mode retries a single block when nothing is found. Confidence is reported per image and words below 80% are highlighted, but confidence is not an accuracy guarantee. Text is editable before copying or export; Word export places one image per page and does not include the images. Generated example checks are not a general accuracy measurement. Images stay in the browser.
Paste a dark-mode screenshot with small text: automatic cleanup inverts and enlarges it before recognition, then copy the corrected text.