To convert a PDF to DXF, open the CADdillo editor, choose File ▸ Import PDF…, pick the page, set the scale with two known points, then File ▸ Save DXF. A vector PDF comes back as editable lines, polylines and text on layers; a scanned PDF has no vectors to extract, so it comes in as a locked underlay to trace over.
Pull real vector geometry out of a PDF drawing, calibrate it to true scale from a dimension you know, and export an editable DXF. The whole thing runs on your own machine — the PDF is never uploaded.
| What you get | Export | |
|---|---|---|
| Vector PDF (plotted from CAD) | Lines, polylines and text at true coordinates, on layers rebuilt from stroke colour and weight | DXF |
| Scanned PDF (a picture of paper) | A locked underlay you calibrate and trace over; nothing is vectorised | DXF of what you trace; the underlay stays out |
Import reads up to 200,000 objects per page and asks first above 50,000; the page picker shows thumbnails for the first 40 pages.
Checked against the converter on .
Vector PDFs give you real geometry. Scanned PDFs get a locked underlay you calibrate to scale and trace — we will not invent lines that were never in the file.
Everything about this conversion depends on that one question, and it is worth thirty seconds to answer it before you start.
Lines stay crisp at any zoom and text is selectable. The drawing operations are still in the file, so extraction recovers genuine geometry with accurate coordinates. This is the good case and it is more common than people expect — any PDF printed straight from CAD software is vector.
Lines blur into pixels when you zoom and no text selects. There is no geometry to recover, so CADdillo brings the page in as a locked underlay you calibrate to scale and draw over. You get real geometry at the end — you just draw it rather than pretending software found it.
The detection is automatic on import. You do not need to classify the file yourself — but knowing which one you have sets the right expectation for how long the job takes.
Plotting to PDF is a lossy, one-way step — it throws away the CAD structure and keeps only what is needed to draw the page. No converter can restore information the file does not contain.
Same extraction, native DWG on the way out.
The reverse trip — free, one click, print-ready sheet.
Extract the outline and cut it on a CNC router.
Starting from photos rather than a PDF? imageinseconds.com converts images into a single multi-page PDF in the browser, free, nothing uploaded.
Open app.caddillo.com, choose File ▸ Import PDF, pick the page, calibrate the scale against a dimension you know, then choose File ▸ Save DXF. The whole conversion runs in your browser — the PDF is parsed locally and the DXF is written locally, so nothing is uploaded to any server at any point.
Not as vector geometry, and CADdillo will tell you so rather than pretending otherwise. A scanned drawing is a photograph of paper — there are no lines in the file, only pixels, and no software can recover geometry that was never there. What happens instead is that the page is brought in as a locked background underlay that you calibrate to scale with two known points, so you can trace over it with real CAD tools and end up with genuine geometry you can trust. That is slower than a magic button, and it is the only honest option. Tools that claim to vectorise a scan are running image tracing, which produces thousands of approximate polyline segments where you wanted a line.
Open it and zoom in hard. If the lines stay crisp at 800% it is vector and this conversion will extract real geometry. If they turn into blurry pixels it is a scan. A quicker test: try selecting text with the cursor — if text highlights, the PDF was generated from software rather than scanned. CADdillo detects this automatically on import and routes the page to whichever path is appropriate, so you do not have to decide in advance.
Because a PDF records paper dimensions, not real ones. It knows a line is 3.2 inches long on a sheet of paper; it does not know that sheet is a floor plan at 1/4in = 1ft and the line is a 12 ft 9 in wall. Nothing in the file bridges that gap. Two clicks and a known dimension does, and it is more reliable than trusting a title-block scale note — those are wrong more often than anyone likes to admit, especially after a drawing has been printed to fit a different paper size.
For a vector PDF, the geometry is as accurate as what was plotted — coordinates are read from the PDF’s own drawing operations rather than inferred from an image. The realistic caveats are that curves are flattened to polylines at a tight chord tolerance, and that any error in your calibration measurement scales the entire drawing proportionally. Calibrating across the longest known dimension you can find minimises that, because a small clicking error over a long span is a smaller percentage.
Yes — same import, then File ▸ Export DWG. That writes a native r2000 DWG and needs a free account, since every DWG conversion runs on our servers. The PDF to DWG page covers that route and its limits.
You will get the geometry back, not the drawing. Plotting to PDF is one-way: it discards layer names, block definitions, associative dimensions and everything else that made the file a CAD model, keeping only the marks needed to draw the page. Extraction recovers linework and text at correct coordinates, with layers rebuilt from stroke colour and line weight and named PDF-C1, PDF-C2 and so on. That is a reconstruction, not the original layer scheme, and it saves hours of redrawing.
Lines and polylines (curves are flattened to a tight tolerance), plus text runs, each placed on a layer named for its stroke colour and line weight. Adjacent line fragments are chained into polylines so the file is not a pile of two-point segments. A scanned page, which is placed as an underlay, is not written to the DXF; only the geometry you trace over it is.
Import reads up to 200,000 objects from the page you pick, and asks for confirmation first above 50,000 because a page that dense can be slow. Multi-page files show page thumbnails for the first 40 pages, and you convert one page at a time.
