Ant Group’s inclusionAI open-sources Ming-Image-0.1-Design (MIT), a 6B text-to-image model for UI, posters and infographics with transparent output, plus a layer-splitting sibling

Your AI and this story

Key facts

• Model card: ‘a 6B text-to-image model for UI, infographics, posters, and other text-rich visual designs’; BF16 safetensors, diffusers and vLLM-Omni support; MIT license
• Recommended: 2048×2048 (1024×1024 for speed), 12 sampling steps, CFG 1.0; the only validated setup is one GPU with at least 80 GiB of memory
• Architecture (ComfyUI Wiki): 30-layer, 3840-dim diffusion transformer with a separate vision-language encoder; Kijai describes the base as Z-Image-derived with a new text encoder; the official pipeline rewrites prompts with Ling-3.0-flash-VL or a Qwen model first
• Design-Layer: separate 6B checkpoint that turns a flattened design plus a layer plan into the requested number of RGBA layers; paired with inclusionAI’s ‘Ling UI Design’ and ‘image-to-editable-PPT’ agent skills in ling-cookbook
• Dates: Hugging Face repo created Sept 17, 2026; DataNorth gives Sept 22 as the release; ComfyUI core support merged Sept 24; repo still updated Oct 14 (425 likes; 3 adapters, 17 fine-tunes, 11 quantizations on the Hub)
• Benchmarks: placement on the Artificial Analysis UI/UX Design leaderboard and Crello layer-recovery results appear only as chart images, no numbers in text; community notes (ComfyUI Wiki) report text failures remain common at INT8