Nimbus 3: Open-source multimodal model tailored for product designers

AI · 6 min read

Nimbus 3 arrives as a community-driven multimodal model trained on interface screenshots, vector art and documentation to produce outputs that understand layout semantics. Unlike generic image generators, Nimbus 3 emits structured layers and bounding boxes so designers can map model outputs directly into design tools.

The model supports tokenized style prompts such as "compact dashboard — 3 columns — left nav" and returns both raster previews and JSON layout descriptors. Early benchmarks show it can produce usable wireframes in under a second on mid-tier GPUs and scale to higher fidelity with additional conditioning.

Developers released a reference plugin for popular design tools and a converter that turns Nimbus JSON into Figma components or SVG layers. The open-source license encourages fine-tuning for niche domains like automotive UIs and accessibility-aware templates, which many studios are already experimenting with.