PNG to DOCX first transcribes the image to Markdown with Deep Read, then our document engine wraps that transcript into an editable Word document.
From real runs
| Element | Result | What happens |
|---|---|---|
| Editable text | Kept | The screenshot words became selectable Word text rather than remaining a pasted image inside the DOCX. |
| Heading | Changed | The large title became a Word heading through the Markdown-to-DOCX step, not a matched screenshot font. |
| Bullets | Changed | The three bullet lines became separate list-like paragraphs; the original bullet glyph and spacing were not reproduced. |
| Callout labels | Changed | Owner, Due, and Status survived, but the extracted DOCX XML showed those soft-break labels running together. |
| Screenshot styling | Lost | The white card, blue callout fill, border, and screenshot dimensions were not preserved as Word layout objects. |
| Images | Lost | The source PNG was not embedded as a page image because this route creates an editable transcript document. |
| Runtime | Kept | The sample DOCX was 10.7 KB and returned in 4.1 seconds after one Deep Read page charge. |
The screenshot words became selectable Word text rather than remaining a pasted image inside the DOCX.
The large title became a Word heading through the Markdown-to-DOCX step, not a matched screenshot font.
The three bullet lines became separate list-like paragraphs; the original bullet glyph and spacing were not reproduced.
Owner, Due, and Status survived, but the extracted DOCX XML showed those soft-break labels running together.
The white card, blue callout fill, border, and screenshot dimensions were not preserved as Word layout objects.
The source PNG was not embedded as a page image because this route creates an editable transcript document.
The sample DOCX was 10.7 KB and returned in 4.1 seconds after one Deep Read page charge.
We converted our PNG sample (219 KB) with AI Deep Read under AI Deep Read on 2026-09-11. It took 3.4 s and produced a 11 KB Word DOCX file, 95% smaller than the input. The engine raised 1 warning, listed below.
Input · sample.png · 219 KB
1240 × 1754 px, 8-bit, colour type 2
Output · sample.docx · 11 KB
Word document: 23 paragraphs, 3 headings, 1 table, 0 embedded images; fonts Consolas
A PNG page is always scanned from the converter's point of view. Deep Read costs 1 credit per image, then DOCX creation happens from the OCR transcript, so typography, card backgrounds, and exact page geometry are not recovered.
Tesseract + Pandoc
tesseract screenshot.png stdout --psm 6 > screenshot.txt
pandoc screenshot.txt -o screenshot.docxThis creates an editable DOCX from OCR text, but it does not infer Word tables or heading hierarchy as consistently as Deep Read.
The HTTP API and the MCP server run the exact engine this page uses (AI Deep Read); the response carries the same warnings the web converter shows. Check support first with GET /api/v1/pairs?from=png&to=docx.
curl -X POST https://filetypeconverters.com/api/v1/convert \
-H "Authorization: Bearer $FTC_API_KEY" \
-F "file=@input.png" -F "from=png" -F "to=docx" \
-F "deepRead=true" \
--output output.docxNo. The tested DOCX contained reconstructed text, not the source PNG as a background. Keep the original image separately if visual evidence matters.
The owner, due date, and status labels came through as soft-break text in the generated document XML. Add tabs or a small table in Word when those labels need a clean handoff.
No. The DOCX uses default document styles produced by our document engine from Markdown. Apply your Word template after conversion if brand typography matters.