Skip to content
Menu

Structured visual data · SVG-native · Commercial AI licence

Structured visual data for AI.

3.29M vector assets with live text, geometry and metadata, licensed for training.

unique assets shipping
3,291,585
datasets, all sampled
4
no outlined type
Live text
Jazz Night A2 Poster — Bauhaus Geometric No.03, structure layer

Every shape as its path geometry

Jazz Night A2 Poster — Bauhaus Geometric No.03

OV553820_poster_bauhaus · 2480×3508 · 70 KB

<text>
10
<path>
11
<g>
3
Live text nodes
Text nodesizewt
arrive from seven68400
sets340600
horn340600
A quartet, a horn section, and no set list after60400
midnight.60400
Doors at eight52400
Two sets, one encore52400
Standing room only52400
bring the family60600
sat 26 sep · the basement, 44 high st48400

Palette declared in file

  • #521f2e
  • #fdeef0
  • #d84868
  • #f8dde2
  • #a88a48

Showing file 1 of 5: Jazz Night A2 Poster — Bauhaus Geometric No.03

fig. 01OV553820_poster_bauhaus.svg from the Poster and Print Dataset · 2480×3508 · 10 text · 11 path · 70 KB · 1 of 5

Why vector

Not just images. Structured visual data.

A bitmap keeps appearance. An SVG keeps the text as text, the shapes as paths, the layout as coordinates and the colour as declared values. Everything a model needs to learn how a design is built is still in the file. The public SVG corpora are icons and emoji; SVGZO's are full pages.

Text
Live <text> nodes with font, weight, size and position
raster: Pixels; OCR required
Shapes
Paths with explicit Bézier geometry
raster: Pixels
Layout
Coordinates, groups, z-order, viewBox
raster: Implicit
Colour
Declared fills and gradients; palette recoverable
raster: Sampled per pixel
How SVG structure trains better models →

From the files

Twelve assets from three datasets, rendered from source. Hover for structure.

Deterministic resvg renders. Raw SVG never reaches the browser from this origin; the sample pack has the originals.

fig. 02–13infographics A2_050001_support_Ullswater-Group_2022, 777013, 633013, IB_IN_000190_lead_wide_teagarden_tabloid_p · posters 123471, OV205856, OV116460, OV243872 · slides 1032000, 1032016, 1032008, 1032057 · every file shown carries live text and zero raster

Presentation slides · 500,000

Slides whose headline is derived from their own chart.

Half a million 16:9 pages composed from PowerPoint template families. Each slide leads with a conclusion, and every figure on it is checked against the story it was drawn from. Chart families are recorded in slot order, so a model can learn which chart carries which claim.

deck template families
306
chart families
131
distinct stories
156,169
16:9, live text
1962×1104
Slide: Core platform takes 42% of the total, structure layer

Every shape as its path geometry

1032010_product_Ottershaw-Group_2017

1032010_product_Ottershaw-Group_2017 · 1962×1104 · 61 KB

<text>
44
<path>
156
<g>
121
Live text nodes
Text nodesizewt
A QUARTER OF TRADING768700
Core platform takes 42% of the total1045700
Ottershaw Group by quarter, 2017 · read in releases672600
4404032700
releases864700
Everything counted this quarter, with the787400
comparison the review turns on kept on the same787400
page as the total.787400
+ 32 more

Palette declared in file

  • #2a3340
  • #048b7e
  • #631330
  • #04a0a9
  • #038087
  • #893e50

Showing file 1 of 8: 1032010_product_Ottershaw-Group_2017

fig. 141032010_product_Ottershaw-Group_2017.svg · 1962×1104 · 44 text · 156 path · 61 KB · 1 of 8 · deck template families from the slide-deck line

Quality you can verify

Every dataset exposes its structure, provenance and licence before purchase.

The six checks every line is measured against, defined on the quality page.

  • SVG Validated
  • Live Text
  • Duplicate Checked
  • Metadata Complete
  • Provenance Documented
  • Licence Verified
byte-identical pages
0
of 850,000 audited, 3 Sep 2026
repeated headline and subject pairs
0
same audit
external references
0
every file is self-contained

See the data before you license it: original SVG files chosen across the production lines, previews, per-file metadata as JSONL and CSV, font notices, README, licence text and checksums.

Licensing

Rights written for model training

Research

For
Academic and non-commercial experimentation on subsets of 25,000 to 100,000 assets.
Train, fine-tune, evaluate
Yes, non-commercial
Model and outputs
Licensee owns
Raw data redistribution
No
Pricing
Per subset from 25,000 assets

Commercial AI

For
Model training and commercial AI products on a sub-line or a full dataset.
Train, fine-tune, evaluate
Yes, commercial models and products
Model and outputs
Licensee owns
Raw data redistribution
No; contractors under NDA
Pricing
Request pricing

Enterprise

For
Custom volume, exclusivity, provenance audit, indemnity and delivery terms.
Train, fine-tune, evaluate
Yes
Model and outputs
Licensee owns, plus IP indemnity
Raw data redistribution
Affiliates and named contractors
Pricing
Custom

Drafts pending counsel review. Full comparison →

Enterprise

Need millions of visual training examples?

From 3.29M ready-to-license assets to custom datasets built around your model's requirements: taxonomy, cleaning, transformation, private delivery, exclusive licensing.

Marketplace

Buy datasets. Sell datasets.

Approved data owners will license SVG, icon, diagram, CAD and animation datasets through SVGZO once quality, licensing and payout standards are live.

Join the seller waitlist →