LLaDA-Image sets open-source SOTA on Qwen-Image-Bench with a 6B DiT and fully released recipes.

A 6B Diffusion Transformer trained from scratch paired with a frozen VLM understanding module on the LLaDA2.0-Mini backbone, building a visual generative prior through image-only pre-training and mid-training across a 220M-sample pipeline before touching paired image-text data. Parameter-free RMSNorm throughout the DiT with the Muon optimizer. The distilled Turbo variant runs inference in 2-4 sampling steps. 53.53 English and 53.38 Chinese on Qwen-Image-Bench, with weights, training code and recipes released. arXiv 2609.03796