Product · Dataset

A dataset that teaches models how to design.

Paired visual-design data for training models that produce structured, editable documents. Each design connects its rendered input to JSON, MarkDesign IR, and HTML outputs.

Get dataset sample

Input-output pairsToken-efficient IRStructured outputs

Dataset scale

3M+

Templates

Presentations, posters, flyers, social graphics, documents, cards, infographics, and more.

7M+

Design pages

More than seven million individual design pages.

100+

Languages

Approximately 67% English, 26% Chinese, and 6% distributed across other languages.

1.9K+

Font families

1,922 active families in the current StarryKit Font Registry, covering Latin, CJK, Arabic, Indic, and more.

Sample demo

Pick a design. Inspect every representation.

Input

Stage source · Full 10-page input · Output: page 1 (Startup) · Page Model 1.5.0

Markdown

Keithston and Partners

Startup

Funding Pitch

Presenting Growth and Opportunity

Presented By

Website

Anna Katrina Marchesi

www.reallygreatsite.com

22 July, 2044

---

Keithston and Partners

What we aim to achieve

INTRODUCTION

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam.

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur.

The vision behind our startup

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam.

---

Keithston and Partners

THE PROBLEM

The gap in the market

Why current solutions are not enough

Common issue faced by people

---

Keithston and Partners

OUR SOLUTION

Simple and practical to use

Designed to grow with the market

Creates real benefits for customers

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam.

---

Keithston and Partners

Size of the potential market

MARKET

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam.

OPPORTUNITY

Growing demand in the industry

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam.

Room for long-term expansion

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam.

---

Keithston and Partners

BUSINESS

MODEL

How we generate revenue

Sustainable growth strategy

---

Keithston and Partners

TRACTION

Supportive response from users

Positive growth shown in early stages

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam.

---

Keithston and Partners

FINANCIAL PROJECTION

Clear plan for managing costs efficiently

Attractive return potential for investors

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua.

Revenue expected to increase steadily each year

Profitability target within the next few years

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua.

---

Keithston and Partners

CLOSING

---

Keithston and Partners

THANK YOU

For Your Attention

123-456-7890

www.reallygreatsite.com

hello@reallygreatsite.com

Design System Prompt

# Design System

## Common Visual Core
Use a 16:9 presentation system built on a deep navy-to-blue linear gradient background, transitioning from near-black navy (#000000) at the left/top toward royal blue (#003271) at the right/bottom. Add bright cyan-blue accents around #1685E5 and #65C5FA, white primary text (#FFFFFF), and cool light-gray body text (#D0D0D0). Use Chocolate Classical Sans as the primary typeface throughout; preserve its clean sans-serif character, with bold weights for headings and labels. Keep the recurring small cyan-and-white brand mark area at the upper left as a replaceable identity element, without depending on any specific source brand.

Headings are large, uppercase or title-case, bold, and white. Cover headlines use approximately 143 px scale with tight two-line stacking; section headlines use approximately 83 px; supporting headings use approximately 29 px; body copy uses approximately 25 px on major narrative pages and approximately 21 px in compact cards. Small metadata and eyebrow labels use approximately 21–30 px. Maintain strong contrast and generous dark negative space.

The visual language combines editorial pitch-deck clarity with rounded geometric accents: cyan-blue pill bars, rounded photo masks, outlined rounded cards, vertical or corner blue capsule forms, circular blue discs, and small grids of white outlined dots. Photography is documentary and business-oriented, used as cropped panels, background overlays, or card imagery with restrained opacity. Keep images secondary to the narrative hierarchy.

## Visual Calibration
Anchor the main content to a broad left margin near 10% of the canvas, with a consistent upper-left identity line and ample vertical breathing room. Use a modular spacing rhythm based on rounded cards, narrow accent bars, and repeated vertical intervals of roughly 75–100 px. Keep headings aligned to a clear column grid. Body copy should remain short, left-aligned, and comfortably separated from headings. Use rounded corners and thin white outlines for feature cards; use cyan fills for emphasis and white or light-gray text inside.

For dark photo covers, apply a navy/black overlay so white typography remains dominant. For image-led layouts, preserve clear subject cropping and avoid placing body copy over high-detail areas. Use cyan accents as directional markers, section dividers, bullet pills, numbered badges, and contact chips rather than as large unstructured color fields.

## Page Families
1. **Cover and closing pages:** Full-bleed or half-bleed business photography under a dark navy overlay, oversized white title treatment, cyan accent capsule or pill, small metadata/contact fields, and decorative dot grids.
2. **Section opener pages:** Large white section title on one side, concise gray narrative copy, and one or more photographic panels or framed image areas. Blue rounded geometric forms may intersect page edges.
3. **Problem, opportunity, traction, and business-model pages:** Split compositions pairing a large heading and narrative text with a photo or stacked evidence blocks. Use cyan arrow pills or circular markers to introduce supporting points.
4. **Solution feature pages:** Centered heading and introductory copy above three evenly spaced outlined cards. Each card has a cyan or blue numbered badge, a short white subheading, and compact gray description.
5. **Market and evidence pages:** Two-column or asymmetrical layouts with stacked portrait-oriented photos on one side and repeated icon-led evidence statements on the other. Keep each statement in a consistent vertical rhythm.
6. **Financial projection pages:** Centered title above a four-card 2x2 matrix. Use blue-filled rounded cards with white headings and small gray explanatory copy, plus a larger supporting image panel where present.
7. **Contact and thank-you pages:** Oversized closing message, darkened background image, rounded cyan contact chips, and a prominent supporting photo or chart image.

## Cross-Family Rhythm
Repeat the upper-left identity position, dark gradient foundation, white heading hierarchy, cyan accent vocabulary, and small dot-grid motif across all families. Alternate image-heavy and information-heavy pages to create presentation pacing. Preserve a consistent left content edge and use a shared baseline for section titles. Keep cards, pills, and image masks visibly related through rounded geometry and restrained outlines.

## Media Logic
Use editable photo placeholders for team collaboration, meetings, strategy, finance, analytics, customer, and workplace imagery. Prefer landscape crops for banners and cover panels, portrait crops for stacked evidence cards, and square or near-square crops for compact feature tiles. Apply dark overlays only when text sits above imagery. Images should support the narrative role of each page family and remain replaceable without changing the grid.

## Adaptation Boundaries
Do not replace the dark navy/royal-blue foundation with light backgrounds unless creating a deliberate alternate variant. Retain white primary typography, cool gray body copy, cyan-blue accents, rounded geometry, and the editorial pitch-deck structure. Keep the page ratio at 16:9 and preserve the hierarchy between oversized section titles, medium supporting headings, compact body copy, and repeated card modules. Source copy, example names, websites, dates, logos, and contact details are editable content only and must not be treated as reusable branding.

Output

Preview

Modern Blue Startup Pitch Deck template from the StarryKit Stage catalog

Inspect the rendered visual result produced from the Markdown content and Design Prompt.

Model training

Build design intelligence across the model lifecycle.

Tokenizer Extension

Extend the base vocabulary with atomic tokens for design primitives, hierarchy, geometry, and style.

Continued Pre-training (CPT)

Adapt the base model to design syntax and structure while preserving its natural-language capabilities.

Grammar-Constrained Decoding (GCD)

Restrict each generated token to the MarkDesign grammar so the model produces valid, renderable structures.

Supervised Fine-Tuning (SFT)

Train on curated Markdown, Design Prompt, and structured-design pairs to improve composition and visual judgment.

Dataset access

Get the sample dataset.

Leave your email and we’ll send a link to download, inspect, and use the sample, including JSON, MarkDesign IR, and HTML.