Skip to content

For machine learning teams

Training data lives on SVGZO

The illustration packs on this site are drawings for a designer to use. Training a model on visual data is a different job under a different licence, so we built SVGZO for it — same company, four dataset lines in the millions, written for model training specifically.

Why it is a separate product

  • Live text and structure, not flattened art

    Every file keeps real text nodes, paths and grouping, so a model can learn layout and typography rather than pixels.

  • Metadata per asset

    Descriptions, tags and geometry ship alongside the files, with Croissant metadata for the datasets that have it.

  • Licensed for training

    Research, commercial-AI and enterprise licences, written for model training specifically. The pack licence on this site does not cover it.

  • Samples before you commit

    Evaluation packs of a few hundred assets, so you can test the data against your pipeline first.

Which one do you want?

Making something — a campaign, a book, an app, a deck
The illustration packs here. 5,321 hand-drawn files in editable SVG and transparent PNG, under one commercial licence. See the packs.
Training, fine-tuning or evaluating a model
SVGZO. The pack licence on this site explicitly does not cover model training, so this is the route rather than a workaround. Go to SVGZO.
Something neither covers
Exclusivity, a custom corpus, provenance audit or indemnity: write to hello@thedoodledesk.com and we will route it.