How it works

From photograph to real-scale CAD drawing

CVCAD replaces the slowest stage of a survey — manually tracing over the raster image — with automatic geometry extraction, while keeping metric control in the technician's hands.

Upload the photo

Start from a rectified photoplan, a photogrammetric orthophoto, or a single frontal photograph, even shot with a smartphone. JPG, PNG and WebP files up to 40 MB are accepted. The more frontal and evenly lit the shot, the cleaner and more metrically reliable the output.

Generate the vector

The computer vision engine analyses the image and derives a high-resolution line-art: edges, material discontinuities, arrises and contact lines are recognised and cleaned of photographic noise. The line-art is then vectorised into continuous paths. Typically 60–120 seconds, around 90 on average. High-resolution photoplans and highly articulated elevations can take longer.

Calibrate the scale and download

On the navigable vector preview you select two points — with automatic snapping to vertices — and enter the real-world distance in millimetres, centimetres or metres. The drawing is regenerated by the backend already to scale, and you download the archive with SVG and DXF: editable vector entities, ready to organise by layer, dimension and integrate into your drawings in AutoCAD, Rhino, QGIS, Illustrator or Inkscape. The DXF is never manipulated after download, so it stays structurally valid.

Metric accuracy and the technician's responsibility

The dimensional accuracy of the drawing depends on the geometric quality of the source image. On a correctly rectified photoplan, two-point calibration returns consistent measurements across the whole surface; on a photograph with residual perspective distortion, measurements remain reliable mainly along the calibration direction. As with any technical deliverable, the metric check and the final validation of the drawing remain the responsibility of the professional who signs it off.

Nature and limits of the deliverable

Nature of the result

Generation is not deterministic. The pipeline includes an interpretive stage in which the model recognises and ranks the geometric entities present in the image: two runs on the same photograph can therefore return non-identical linework, differing in the density of detail and in which elements are returned. The behaviour is comparable to two technicians independently tracing the same photographic survey: both produce a correct deliverable, neither identical to the other. This is not a malfunction but a characteristic of the process.

Effect of resolution

The level of detail returned depends on the actual resolution of the source image, not on its nominal pixel dimensions. Low-resolution, resampled or aggressively compressed images produce simplified linework, with minor profiles lost. On noisy images the vector output can contain short, irrelevant segments generated from colour variations with no geometric counterpart.

Frame proportions

Processing works on any format. Rendering is nonetheless more uniform, and the likelihood of artefacts lower, on frames with contained proportions: a 1:1 ratio offers the best conditions, followed by 4:3 and 3:2. As the frame becomes more elongated, detail is spread over a larger extent and rendering tends to differ between the central portion and the edges. On markedly panoramic frames it is preferable to work in separate portions, as described in the next point.

Extended elevations

For fronts of considerable length or high decorative density, we recommend splitting the elevation into separate portions, each processed independently and recomposed in the CAD environment. This approach keeps the ratio between resolution and surveyed surface constant, and returns a higher level of detail than processing a single overview image. Portions should be shot with enough overlap to allow alignment on recognisable elements.

Checking the deliverable

The credit is charged at the moment of generation. The resulting vector preview allows you to check the linework and calibrate the scale before downloading, which incurs no further charge: SVG and DXF remain downloadable at no extra cost. Deliverable quality depends decisively on the source image, for which we recommend the shooting criteria in the dedicated guide.

The output is a vector base, not a final deliverable. It requires metric and formal verification before use in technical documentation: checking the calibration against a known distance independent from the one used for scaling, removing irrelevant entities, and completing any unresolved portions. Responsibility for the final deliverable rests with the technician in charge.

Try CVCAD on your next survey

Open the app