Image in, real SVG out
Every asset is a vector source that can be rendered deterministically to raster at any size — so every file is an aligned (image, ground-truth SVG) pair with clean geometry, not a trace.
KOLCO CLUB · Training data for design models
3,096,618 editable SVG design files, shipping today. Every one carries live text, its own metadata row and a documented origin — and no generative model made any part of any of them. Licensed to AI companies for training, and open to creators who want their work in it.
Nobody wakes up wanting poster data. They wake up because a model cannot render legible text, or cannot produce an editable vector, or cannot lay out a page. These are the same three million files, cut along the axis of what they teach.
Every asset is a vector source that can be rendered deterministically to raster at any size — so every file is an aligned (image, ground-truth SVG) pair with clean geometry, not a trace.
Text rendering is a known weak spot, and scraped data has no ground truth for it. Here every string is a live SVG <text> node with its exact content, position, size and colour — readable before a pixel is drawn.
Every text node carries an explicit font size and position, and every page uses more than one size — so typographic hierarchy and reading order are recoverable per file, across 23 print formats and 587 layout families.
On any infographic the share breakdown, ranking, trend and comparison are four views of one set of numbers. A page is internally consistent before you change it, and still consistent after — which is what makes it usable for chart reasoning rather than chart mimicry.
“Derivable” means exactly that. The pairs, the text boxes and the hierarchy are all present in the files and extractable by a deterministic pass — we have verified that on the public sample and you can too. They are not pre-packaged SKUs sitting on a shelf. Tell us which one you need and we will cut it against your schema.
Shipping collections from The Doodle Desk, not a roadmap. Every file is real SVG with live editable text — no outlined type, no flattened artwork — and every asset carries its own metadata row, so a lab can license a slice instead of running a scrape.
Single-page print products at known trim sizes — the headline, body copy, call to action and contact line all live text.
Charts, rankings and breakdowns where the arithmetic on a page agrees with itself — before and after you change a figure.
Studio to four bedroom, 237 to 2,186 sq ft, on 20 standard CAD layers — walls, doors, glazing, columns, furniture and annotation kept separate. A spatial-reasoning corpus rather than a design one, so it has its own page.
Every figure below was run against the 191 SVG files in the public review sample — stratified, drawn at random within each format, nothing retouched. Download it and reproduce the table yourself.
| Check | Posters | Infographics | Floor plans |
|---|---|---|---|
| Files in sample | 92 | 48 | 51 |
| Parse as valid XML | 92 / 92 | 48 / 48 | 51 / 51 |
| Carry live, non-empty <text> | 92 / 92 | 48 / 48 | 51 / 51 |
| Reference a remote font or asset | 0 | 0 | 0 |
| Median file size | 88 KB | 76 KB | 41 KB |
| Median live text nodes | 7 | 61 | 151 |
| Median vector paths | 20 | 32 | 275 |
Fonts travel inside the file as data URIs, or are named system faces with generic fallbacks. Nothing reaches outside itself for a font or an image, so a file renders identically wherever it lands — including inside a training pipeline with no network. The full breakdown — class distributions, the label schema, worked examples and the provenance record — is in the datasheet.
191 files with their metadata indexes, catalogue indexes and licence notes. No form, no email, no call. Open them in any vector editor and judge them.
These files are for evaluation, not redistribution. The commercial terms travel with the full pack — ask and we will send the licence before any conversation about price.
Evaluate for free, then licence what you need. Scoping goes down to the sub-collection — icon sets alone, or floor plans alone — so you are never obliged to take the whole catalogue to get the part you need.
The public sample plus, on request, a larger evaluation slice cut to your taxonomy. For assessment only — no training grant.
A versioned release of one or more collections, licensed for model training, with the documentation your model card needs. Need something the catalogue does not hold? That gets drawn to your brief under the same terms.
Your S3 or GCS bucket, a private Hugging Face repo, or signed archive links.
WebDataset shards or Parquet, sources and renders addressed by checksum.
Immutable release tags. Additions ship as new tags; withdrawals are listed.
Consent manifest, QA report and a provenance summary written for a model card.
The catalogue today is our own production. We are opening it to other designers: submit original vector work, we check and label it, and it joins the releases that AI companies licence. You are paid a share of what each release earns, for as long as it earns.
No. You keep full ownership. What you grant is a non-exclusive licence for the work to be included in dataset releases used for model training. You can keep selling, printing and licensing the same files anywhere else, at the same time.
Each release earns licence revenue. A fixed share of that revenue is distributed across the creators whose files are in the release, weighted by how many of their accepted files it contains. Payouts run monthly, and every line is traceable to a release and a buyer.
Yes, for anything that has not shipped. Withdrawing removes your files from all future releases and refreshes. Releases already licensed remain valid for their term, because a buyer has already trained on them — that limit is stated plainly in the grant you sign, before you submit.
Yes. Scoping goes down to the sub-collection — icon sets alone, or floor plans alone, or festival illustration alone. You are not obliged to take the whole catalogue to get the part you need.
Yes, and you should. A 191-file review sample is published for the three shipping collections — stratified rather than cherry-picked, drawn at random within each format, with nothing retouched. It ships with the per-file metadata index, the catalogue index and the licence note, so you can measure the claims on this page yourself before a contract exists.
No. No generative model produced any part of any file in the shipping packs, and the licence note that travels with each pack states it. That matters more than it used to: a design model trained on another model’s output inherits its artefacts, and there is no way to detect that after the fact from the files alone.
Get in touch