Deals · Enterprise AI

F13 raises $5M Pre-Seed to build AI that generates precise, editable vector graphics for professional use

Develops foundation models for generating precise, technically accurate and fully editable vector graphics from text, images or existing graphics — targeting professional workflows such as data visualisations, infographics, scientific diagrams and brand-consistent layouts.

ProYarn Desk · September 23, 2026Read this in French

in 𝕏 ✆

“We didn't want to build another model that's impressively expressive and quietly wrong. We create visuals that are not only beautiful to look at, but are true to the brief, with the correct proportions, accurate data, and precise layouts. That's the only version of this technology that's genuinely useful to discerning professionals who know what good looks like and are doing high-stakes design work.”

Gregory Janik — CEO and co-founder, F13

F13 (Berlin) has raised $5 million (approximately €4.7 million) in a Pre-Seed round co-led by Credo Ventures (Prague) and Point Nine Capital (Berlin), with Carles Reina (Baobab Ventures) and Jack Richardson (mainframe) also participating. The company is emerging from stealth with a plan to build foundation models for generating precise, fully editable vector graphics. Founded by Gregory Janik (CEO) and Ingimar Tomasson (CTO).

The problem F13 is solving is specific. AI image generation has become capable enough to produce outputs that look right at a glance but fail the accuracy test that professional use requires: a data visualisation with bars in approximately the correct proportions, an infographic where the numbers are nearly right, a technical diagram that captures the general shape but misses the specification. For casual use, approximately right is fine. For a designer building a client presentation, a consultant producing a board report, or a scientist preparing a paper figure, it is not.

Why vector

The key design choice is the output format. Raster images — pixels arranged in a grid — have no inherent structure. When an AI generates a raster output, it is generating colour values, not semantic elements; correctness constraints (this bar should be exactly 47 units tall; this label should read exactly "Q3 2026") have to be enforced at the generation step, which is difficult and still imperfect. Vector graphics are structured: every element is an explicit path or shape with coordinates, text with editable content, nodes that can be snapped to exact positions. Generating vector graphics means generating structure, which makes the output verifiable and correctable in a way that a pixel grid is not.

The team and the round

Janik is a serial founder with two prior exits; Tomasson is an ML researcher trained in mathematics at UCL. The Pre-Seed round will be used to bring the first model to market, expand compute and data, and hire; a public launch is planned for late 2026.

Point Nine Capital has previously backed Revolut, Attio, and Zendesk — predominantly software and AI companies — and recently backed the Dutch crop science startup Aardaia in a departure from its usual mandate. Here it is on familiar territory. Credo Ventures is one of Central Europe's most active early-stage funds, with a portfolio spanning B2B SaaS and deep-tech. The co-lead structure at Pre-Seed suggests competitive conviction from both sides. The 18-month test is whether F13's accuracy threshold is sufficient to replace, rather than merely reduce, the manual correction step that professional designers currently accept as inevitable.

Why it matters to investors

Current AI image generators break down for professional use in a specific, predictable way: the outputs look plausible but are structurally wrong. A bar chart generated by a diffusion model will have bars in approximately the right proportions but not exactly; an infographic will have approximately the right text but not verbatim; a technical diagram will be roughly correct but not to specification. The workaround — generate an approximate image, export it, then fix it in Illustrator — mostly negates the productivity gain. F13's bet is that vector graphics sidestep this problem: the output format is structured (paths, nodes, layers, anchor points) rather than a pixel array, which makes correctness constraints enforceable in a way they are not for raster images. The public launch later in 2026 is the milestone that will determine whether the accuracy claims hold under real professional workflows. **Point Nine Capital** — which has backed [Aardaia](/en/deals/aardaia) and previously Revolut, Attio, and Zendesk — investing at Pre-Seed stage alongside **Credo Ventures** (Prague) signals conviction in the team rather than proven metrics. The 18-month test is whether professional designers find the accuracy threshold sufficient to reduce rather than merely augment the manual correction step.

Sources

Pre-Seed Enterprise AI Germany Credo Ventures Point Nine Capital Carles Reina Jack Richardson

Threaded to this story

Enterprise AI · July 28, 2026

Every European round, in your inbox by 8am.

The day's seed and Series A rounds across France and Europe — threaded, sourced, and read in two minutes. Free.