Ming-Image-0.1-Design

Ming-Image-0.1-Design is a 6-billion-parameter open-weight text-to-image model from inclusionAI, the Ant Group research lab behind the Ling models, published on Hugging Face under the MIT licence and reported as released on 22 September 2026 (DataNorth, 23 September). It is built for visual design rather than general imagery: complete user-interface screens, dashboards, infographics and posters, including the text inside them, from structured prompts. It can also output RGBA images with a transparent background, switched on by a fixed prompt prefix in English or Chinese. The model card recommends 2048 × 2048 output with 12 sampling steps, and the only configuration validated so far is a single GPU with 80 GB of memory in BF16, served through vLLM-Omni — so running it in-house is a data-centre GPU decision, not a laptop one. There is no hosted API or price from the developer. It was released with a companion layer-splitting model, Ming-Image-0.1-Design-Layer, and two open-source agent skills, one for interface design and one that turns generated slides into editable PowerPoint files. inclusionAI reports first place among open-weight models on Artificial Analysis’s UI/UX design leaderboard, but publishes that only as an image on its model card; it is recorded here as the developer’s own claim, not as a score. For a buyer the MIT licence is the notable part: it allows commercial use without the revenue thresholds that several other open image models carry.