Ming-Image-0.1-Design
Two open 6-billion-parameter models that generate complete visual designs, such as posters and UI mockups, and decompose a flattened design image into separately editable layers.
Ming-Image-0.1-Design generates visual designs, including UI screens, infographics, posters and text-heavy compositions, from a text prompt, at resolutions of 1024 or 2048 pixels by default. Its companion model, Ming-Image-0.1-Design-Layer, takes a flattened design image and decomposes it back into independently editable transparent layers, so elements such as text and background can be moved or edited separately after generation.
The project describes two downstream agent skills built on the models: one that uses the generated visuals and layer decomposition to help an agent build and visually check UI code from a prompt or a screenshot, and another that recreates a generated slide image as an editable PowerPoint slide with text and shapes converted to native elements.
It requires Python 3.10 or newer and CUDA-capable PyTorch, and is licensed under MIT.