A 6B image generator ships with its weights, its code and its recipe

LLaDA-Image pairs a 6B diffusion transformer with a frozen vision-language model. Its weights, code and training recipe ship under CC BY-SA 4.0, against closed pipelines that publish none of the three.