inclusionAI has published Ming-Image-0.1-Design, a 6B text-to-image model aimed at UI, infographics, posters and other designs where text carries the meaning. The model page on huggingface.co describes it as generating complete visual compositions rather than isolated objects, and as supporting RGBA output with transparent backgrounds. The licence is MIT.
What the transparent output changes
Most image models return a flat picture. RGBA output adds an alpha channel, so the background can be empty and the subject can be dropped onto another surface without cutting it out by hand. The page gives one instruction for this: prepend exactly one of the recommended RGBA phrases to the prompt. It also notes that the checkerboard seen in the gallery is only a preview of transparency and is not part of the generated image.
The settings the page recommends
Resolution is 2048 x 2048, or 1024 x 1024 for faster generation, and the public inference code maps a resolution request to one of those two buckets. Sampling steps are 12, CFG scale is 1.0, and precision is BF16. The validated hardware configuration is a single CUDA GPU with 80 GiB of VRAM. For serving, the page points to vLLM-Omni, with recipes and an installation guide.
Installation and inference run through a companion Ming-Image repository on GitHub, with a command-line script that takes a model name, a task, a prompt file, a resolution and an output directory. Prompt enhancement can use Ling-3.0-flash-VL or qwen3.8-27B, described on the page as text-to-image prompt rewriting.
The page also lists a UI/UX design leaderboard, separate design and presentation skills, and a demo. Fifteen finetunes and seven quantizations of the model are listed, along with six Spaces that use it. Downloads are not tracked for this model, so the page shows no download figure.
The concrete detail to hold on to is the alpha channel: a 6B model that returns a composition with an empty background, at 2048 x 2048, under an MIT licence.
Source: huggingface.co — BARGO’s commentary on the linked source.
