Understanding
The model learns how design works and how each brand looks. So what it makes is the brand, not the AI average.
The applied AI research lab for design work
AI treats design as an image generation problem and hands you one flat image.
At Coco Labs, we train our own models to build designs in layers. Our ambition is to make design as fundamental to AI as text and images are today.
Introducing Design Intelligence
Design Intelligence is a new class of model, trained by Coco Labs. It builds every design in layers, from real assets.
Image models generate pixels. Our model builds designs, element by element.
Construction
Proprietary design language that preserves every element's native format and enables export to external design tools such as Figma.
Control
Change one element, and everything else stays as it was. Uploaded photos keep their full resolution.
Vyb - Our First Design Model
Vyb is singularly trained to generate layered, editable design, which keeps it small and fast.
In head-to-head testing against GPT Image 2.5, Muse and Recraft 4.1, Vyb ranked first on brand color, typography and editing.
 Vyb |  GPT Image2.5 |  Muse |  Recraft4.1 | |
|---|---|---|---|---|
| Typeface accuracy | 95.3% | 78.2% | 59.8% | 51.2% |
| Color error | 0.0 | 4.12 | 5.63 | 27.79 |
| Design preserved across edits | 88.7% | 4.1% | 9.3% | 0.8% |
See what Design Intelligence can do
See it on your own brand. Vyb is in private beta.