Ming-Image opens editable design generation

InclusionAI has released two 6B-parameter, MIT-licensed Ming-Image models: Design generates text-rich UI, infographic, and poster compositions, while Design-Layer decomposes flattened designs into editable RGBA layers. Both model cards specify BF16 inference on one 80 GiB GPU; Design recommends 2,048-pixel output and Layer includes a six-layer example. For builders, this is an open design pipeline worth testing locally, not lightweight endpoint economics—the archive’s Muse download analysis puts adjacent consumer-image demand in perspective.