Coco Labs 的 AI 设计模型 Vyb 按图层逐个元素构建设计，文字保持可编辑文本、图形保持矢量、照片保持像素，并从网站自动提取 logo、字体和配色。在 145 款 Google 字体上按像素重叠 IoU 打分，Vyb 达到 90.3%，对比中 最强图像模型为 76.0%，该测试的测量上限为 94.4%。由于每层保留原生格式，导出 Figma 时结构完整保留。
Design is the next place where models are learning to output structure instead of pixels.
Vyb, a design model from Coco Labs, builds a graphic element by element, each on its own layer, so text stays live text, graphics stay vectors and photos stay pixels.
That changes what you get back. The headline can be retyped. The logo is the logo file, pulled from your website along with your fonts and exact colors.
Across 145 randomly chosen Google fonts, scored by pixel overlap (IoU) against a programmatic render, Vyb reached 90.3%. The measurement ceiling on that test is 94.4%, and the strongest image model in the comparison reached 76.0%.
Because every layer keeps its native type, the export to Figma carries the structure across intact.
I launched the first image and video generation models at Google, and kept asking why all AI designs look the same. AI has brought us the WORST era of design. It made content creation effortless, but it also made 90% of it look painfully generic. My co-founder and I spent 2 years building in stealth to change that. Meet Vyb, the first AI design model that understands your taste like a designer would. RT + comment your website URL and we’ll make your first design.在 X 查看被引用的帖子
来源：Rohan Paul · x.com