Ant Group has released the Ming-Image-0.1-Design model series as open source. The collection includes two 6B parameter models: Ming-Image-0.1-Design, which generates UI designs, posters, and infographics from text, and Ming-Image-0.1-Design-Layer, designed to decompose images into editable transparent layers.
The models support 8K structured prompts, allowing users to organize requirements by copy, layout, and visual style. They feature optimized text rendering for UI elements like buttons and cards, and use a native RGBA VAE to generate transparent assets. The company also released Design Skill and PPT Skill to integrate these capabilities into web pages and presentations.
According to an Artificial Analysis evaluation updated on September 18, 2026, Ming-Image-0.1-Design ranks first among open-source models in the UI/UX category with an Elo score of 1082. The model achieved a 67.4% win rate in layout, 67.0% in complex composition, and 66.7% in text rendering.
Ant Group noted current limitations, stating that the models may struggle with complex hand gestures, continuous action sequences, or fine shadow and reflection details. The layer decomposition process is subjective, and complex lighting or occlusion can occasionally lead to artifacts or missing layers.
To facilitate testing, the company is offering free API access to the Ming-Image-0.1-Design series via OpenRouter for a period of two weeks.
