A 6B unified image generation + editing model distilled to 4 sampling steps, driven by a frozen LLaDA2 MoE understanding backbone. English and Chinese prompts, including text rendering.
Paper · Code · Model