Paper drawings, scans and photos (beta)
Paper drawings and scanned images to DXF with AI
AI recognises lines, circles, arcs and text in scans, photos, faxes and scanned PDFs of paper drawings and writes a DXF draft with layers and line types. No CAD, no installation; sign in and use up to 3 pages a day for free. Results belong to the “AI recognition” tier: a human check is expected.
PNG / JPEG / TIFF / PDF (scans, photos, scanned PDFs), up to 20MB. PDF pages are sorted into scanned / vector automatically.
Sign in to use AI recognition (free: one page per run and 3 pages a day; a day pass has the same allowance; Pro: 100 pages a month).
Turning a paper drawing into a DXF
1. Drop a scan, a photo or a scanned PDF into the box above
A JPG, PNG or TIFF image, or a scanned PDF (one page per run). Scans from a copier at 300 dpi or more work best; take photos square-on to the sheet without shadows.
2. Sign in and start AI recognition
Sign in with the link sent to your e-mail address (no password, no registration form). The free tier allows 3 pages a day. Recognition takes from tens of seconds to a few minutes.
3. Check the overlay and download the DXF
The overlay shows the input in grey, recognised lines in red and ink that was not picked up in blue, so gaps are visible at a glance. Open the DXF in your CAD, check and correct it before use.
Which drawings work and which do not
- Scans of paper drawings (JPG, PNG, TIFF; 300 dpi or more recommended)
- Photos of drawings taken with a phone or camera (square-on, no shadows)
- Scanned PDFs (image-only PDFs) and drawings received by fax
- Vector PDFs exported from CAD (where text can be selected) are converted without AI by structure recovery (see how to tell them apart below)
- Hand-drawn drawings can be recognised when the line weight is even, but accuracy drops
- Ellipses, splines and the meaning of dimensions (tolerances, symbols) are not supported; they come through as lines and text
- Text is read by OCR; low-confidence words need checking
How to tell a vector PDF from a scanned PDF
A PDF is either a vector PDF exported from CAD (lines and text stored as data) or a scanned PDF (an image only). Only scanned PDFs need AI recognition.
You can drag to select text in a PDF viewer → vector PDF
Converted without AI by structure recovery to DXF, JWW or DWG: lines, arcs, text and layers are recovered and the scale is estimated from the dimensions so the result is at real size.
Text cannot be selected and lines blur when you zoom in → scanned PDF
Use the AI recognition on this page: lines, circles, arcs and text are read into a DXF draft with layers and line types.
Not sure
Drop the file on the home converter; each page is classified as vector or scan automatically and you are guided from there.
Compared with tracing services and raster-to-vector software (at the time of writing)
| Method | Cost | Time | Result | Best for |
|---|---|---|---|---|
| zumen-gate AI recognition (this page) | Free tier (3 pages a day); Pro 100 pages a month | Minutes | Lines, arcs and text in a layered DXF; a human check and corrections are expected | You need a draft now, or want to try first |
| Tracing service | Roughly US$30–100 per sheet | Days to weeks | Drawn by a person, so accurate; dimensions become CAD entities | Deliverable-grade accuracy |
| Raster-to-vector software | Paid (hundreds of dollars and up) | Immediate | Lines are captured, but text, layers and scale are manual work | Large volumes; you can install software |
| Trace by hand over the image in CAD | Free if you have CAD | Hours per sheet | Accurate, because you draw it yourself | A few sheets and CAD at hand |
What the DXF contains
- Lines, circles and arcs on outline / hidden / centre / dimension / text / frame / hatch layers, with continuous / dashed / dash-dot / dotted line types
- Text read by OCR, written as TEXT entities
- The scale is estimated from dimension values; when enough dimensions agree, coordinates are converted to real-world millimetres and the matching dimensions are written as dimension entities too
- If the scale cannot be estimated, coordinates stay in image pixels (or paper millimetres when the dpi is known); apply a scale factor in your CAD
- The overlay (grey = input, red = recognised, blue = ink not picked up) shows where the result is incomplete
Allowance and pricing
- Sign in to use it (beta). The free tier allows one page per run and 3 pages a day.
- Pro (monthly or yearly) allows 100 pages a month and multi-page PDFs in one run.
- The one-time passes (24 hours, 7 days) do not include AI recognition (same allowance as free).
- Data handling is the same as for normal conversions (the free tier uses images and results to improve the model; Pro does not).
- An MCP server and API for reading, editing and converting drawings from Claude Code and other AI clients are also available (MCP and API page).
AI recognition FAQ
How accurate is it?
Lines, circles, arcs and text are recognised, but faint lines, overlapping lines and handwriting cause gaps and misreads. Use the overlay (grey = input, red = recognised, blue = ink not picked up) to find what is missing and correct the DXF in your CAD. For deliverable-grade accuracy, consider a tracing service or tracing by hand.
Does it work with hand-drawn drawings?
Yes, when the line weight is even and the background is clean and white. Freehand lines come through as polylines and text recognition is weaker. Try one page first and check the overlay.
What happens to the scale and real-world size?
The scale is estimated from the dimension values and extension lines; when enough dimensions agree, coordinates are converted to real-world millimetres. Otherwise they stay in image pixels (or paper millimetres when the dpi is known), so apply a scale factor in your CAD. The result is noted in the loss report.
How many pages are free, and do I need to sign in?
Sign-in is required (a link is sent to your e-mail address; there is no registration form). The free tier allows one page per run and 3 pages a day; Pro (monthly or yearly) allows 100 pages a month and multi-page PDFs in one run. The 24-hour and 7-day passes do not include AI recognition.
Should I upload JPG, PNG or PDF?
Upload the file as it came from the scanner or copier, without recompressing or shrinking it. Use 300 dpi or more; on the free tier multi-page PDFs are processed one page at a time. Photos can be JPG; take them square-on without shadows.
Can I upload confidential drawings?
On the free tier, images and results are used to improve the model (training), so please do not upload confidential drawings. Transfers and storage are encrypted and nothing is shared with third parties. Pro does not use your files for training and deletes them right after conversion. See "How we handle your data" for details.