Specifications

What every SVGZO file is.

A buyer's checklist, answered per dataset. The same rules are enforced by the ingest pipeline that builds previews and sample packs, and every claim below is measured from the files, not asserted.

Explore datasets

Format

SVG 1.1, UTF-8, one document per file, with width, height and viewBox declared.
Self-contained: no external references, no scripts, no foreignObject. Fonts are embedded or referenced by name; embedded fonts carry their authors' notices in NOTICES/.
Renders deterministically. Previews on this site are produced with resvg from the shipped files.
Every file has a SHA-256 checksum, a stable asset id and a metadata row.

DOM nodes

Files are real SVG documents whose Document Object Model nodes determine how the image renders. A model trained on them sees structure, not pixels.

DOM node types
Node	What it carries	Presence
`<text>`	Actual editable text strings: headlines, labels, figures, captions, schedules. Never outlined to paths. Font family, weight and size are declared on the node.	Present in every file of every dataset
`<path>`	Vector layouts and geometry: chart marks, illustration, walls, openings, furniture, decorative shapes. Bézier and line commands with explicit coordinates.	Present in every file of every dataset
`<image>`	Embedded high-resolution raster images, where a design includes a photographic or bitmap element. Recorded per file as image_count so buyers can filter for vector-only subsets.	Absent from most lines; see the table below
`<g>`	Groups that carry layout structure: sections, chart slots, architectural layers (A-WALL, A-DOOR, A-GLAZ, A-FURN, A-ANNO).	Present in every file

Interleaved text and graphics

Text and graphics in every file are made in the context of one another. A headline states what the chart beneath it shows; a legend names the series drawn beside it; a room label sits inside the room it measures; a poster's copy describes the illustration it accompanies. The text is not a caption bolted on after the fact, and the graphics are not decoration unrelated to the words. This is what makes the files usable for text-to-SVG, chart understanding and layout models without a separate annotation pass.

Content rules

No brands or trademarks. Organisation names, product names and logos are invented placeholders. No real company, brand mark or trademarked character appears.
No personally identifiable information. No real person is named, depicted or described. Contact lines on print templates use placeholder patterns such as +00 000 000 0000 and www.yourshop.com. Floor plans describe no real property or address.
Placeholder figures. Every statistic, price and measurement is written for the layout. The data teaches structure, not facts.
No third-party content. Assets are composed, created and processed into SVG from source files by The Doodle Desk's team and network of creators.

Per dataset

Specification compliance per dataset
Dataset	<text> nodes	<path> nodes	<image> nodes	Interleaved text and graphics	Measured on the preview set
Infographic Dataset	Yes	Yes	None. Files carrying an embedded bitmap are excluded from delivery, samples and previews; every delivered file is vector only	Yes	122 files: 8,143 text, 6,838 path, 0 with image
Poster and Print Dataset	Yes	Yes	None. Files carrying an embedded bitmap are excluded from delivery, samples and previews; every delivered file is vector only	Yes	79 files: 757 text, 2,345 path, 0 with image
Presentation Slide Dataset	Yes	Yes	None. Files carrying an embedded bitmap are excluded from delivery, samples and previews; every delivered file is vector only	Yes	120 files: 4,759 text, 7,911 path, 0 with image
Indian Residential Floor Plan Dataset	Yes	Yes	See dataset page	Yes	120 files: 20,219 text, 42,016 path, 0 with image

Image-node counts for the full lines are published on each dataset page. Buyers who need vector-only data can filter on image_count = 0 in the metadata.

Verification you can run

Every sample pack ships metadata with text_count, path_count and image_count per file, SHA256SUMS, and the pre-flight report that excluded any file with raster, script or external references. The same numbers can be reproduced with any XML parser.