Doc 1 · Editorial
Poster · 1200 × 1600
Model · In development
VCM is learning to compose a poster from structured data.
VCM is still in development and is being trained on structured visual-design data. Explore the paired dataset behind it, or try the AI-assisted design workflow available in StarryKit today.
VCM · poster
The pipeline
Markdown InVisually Structured Document Out.
Content streams in on the left. VCM parses it into a structured composition. The visual document streams out on the right — fully layered and editable, not flattened.
content.md
streaming# AFTER IMAGE > Independent Culture Festival · North Dock / Hall 03 ## Programme - Three nights of sound, moving image and print - Live sets · film premieres · small press · night radio - Fri—Sun · 18:00—late ## Dates 08—10 SEP
→
VCM · v1
Composing…
{
"size": { "width": 1200, "height": 1600 },
"content": {
"background": { "type": "solid", "color": "#E8DFCF" },
"nodes": [
{ "id": "coral-field", "type": "shape",
"layout": { "x": 816, "y": 0, "w": 384, "h": 1600 } },
{ "id": "olive-index", "type": "shape",
"layout": { "x": 60, "y": 120, "w": 156, "h": 192 } },
{ "id": "headline", "type": "text", "content": "AFTER\nIMAGE",
"layout": { "x": 96, "y": 368, "w": 1008, "h": 432 } },
{ "id": "description", "type": "text",
"layout": { "x": 108, "y": 928, "w": 648, "h": 208 } },
{ "id": "date", "type": "text",
"layout": { "x": 888, "y": 1104, "w": 228, "h": 112 } },
… 10 more nodes
]
}
}→
after-image.doc
15 layersSame content · different prompts
Your LLM handles the content.VCM handles the design.
Pick a design prompt. VCM recomposes the same source into a cohesive set — every document in the set shares the same visual language.
design prompt
Editorial
Create an editorial cultural-festival system with an oversized high-contrast serif headline, an asymmetric magazine grid, precise sans-serif metadata, and generous margins. Use warm paper, near-black, coral, and muted olive; keep every supplied phrase exact and editable.
Doc 2 · Editorial
Square · 1200 × 1200
Doc 3 · Editorial
Deck · 1920 × 1080
Model · features
What's under the hood.
Model card
VCM · V0.1
A purpose-built model for structured visual composition, trained on structured design, not scraped bitmaps.
14B
Parameters
64K
Context
Purpose-built
Visual Tokenizer
$0.15
/ 1M input tokens
$0.60
/ 1M output tokens
200+
output tokens / second
Target specifications. Final performance and pricing may vary at launch.
Answers
The short version.
Is VCM a stock-image model?+
No. VCM doesn't hallucinate photorealistic pixels. It composes real, editable layers — frames, text, shapes, tokens.
Is it a template engine?+
No. There are no fixed slot layouts. Every document is composed for its content and the design prompt.
Can I plug my own content pipeline in?+
Yes. VCM accepts markdown, plain text, JSON and structured briefs. Output is a structured document any downstream tool can open.
How editable is the output?+
Completely. Every layer is selectable, restylable and re-composable. Nothing is baked into a bitmap.
Does it respect my brand tokens?+
Yes. Load a token set and VCM composes strictly inside it — palette, type ramp, spacing, motion.
Where do I use VCM?+
In StarryKit, our co-editing surface — and through the API for embedding into your own products.
Model status
VCM is still in development.
We are still training and evaluating the model. Explore the dataset behind it, or try the AI-assisted design workflow already available in StarryKit.