Does it upload my notes?
No. Your pages are decoded locally. On first recognition, the browser downloads the publicly hosted runtime and model weights; recognition uses on-device inference.
Turn photographed pages into a structured text draft. Review each line, preserve paragraph breaks, and export your work—without uploading your notes to an OCR service.
Standard / No AI · Import → Prepare → Mark lines → Transcribe → Review → Export
Your handwritten page will appear here.
Add a handwritten note or explore the sample to get started.
Select “Draw missing line” to add a recognition region on the page.
Type manually, edit region boxes, set reading order and mark reviewed. AI is optional.
Type into detected or manually drawn lines to enable export. AI is optional.
No. Your pages are decoded locally. On first recognition, the browser downloads the publicly hosted runtime and model weights; recognition uses on-device inference.
No. The line and paragraph boundaries come from image heuristics and may need correction for columns, tables, ruled paper or unusual spacing.
Clear, English-language handwriting in reasonably straight lines. TrOCR may struggle with dense cursive, math, diagrams and mixed languages.
Image preview, layout editing and exports work offline. Recognition requires the runtime/model to have been downloaded and cached by the browser, or self-hosted.
Preparing your document…
Import page images, detect lines and add missing regions as needed. Recognize selected blank lines with optional TrOCR, proofread the text and paragraph boundaries, then export before leaving. Overwriting corrected text requires the explicit replacement option.
TrOCR is intended for English handwritten single lines, not multilingual pages, equations or tables. Line grouping is heuristic and recognition needs proofreading. External runtime/model downloads are required for AI. Workspace/page/crop budgets apply. Use Save editable project to download a portable JSON containing the working page images and transcriptions. TXT/Markdown/HTML/transcription JSON remain text-only exports. PDF/DOCX import is unsupported.
References: TrOCR handwritten model · Transformers.js
Provide a clear image of a handwritten line for TrOCR recognition, then review or correct the result. For paragraphs, use the available heuristic line/paragraph review workflow; it does not replace manual proofreading.
Authored example: Write “Bring the blue folder on Friday.” by hand, photograph it in even light, and compare the recognized sentence character by character. Try a paragraph only after checking how the line detector handles your layout.
Limits and model handling: Transformers.js and the TrOCR model download when recognition is requested. The supplied model targets English handwritten single lines; detection for paragraphs is heuristic. PDF/DOCX import, multilingual or equation recognition, and persistent drafts are unsupported.
Page and paragraph line detection is heuristic, and recognition is intended for English handwritten single lines. Review and correct the result.
No. Exported text formats omit the source images; retain the originals separately.