Hugging Face researchers have introduced LLaDA-Image, a unified framework that combines a 6-billion-parameter Diffusion Transformer (DiT) trained from scratch with a frozen vision-language understanding module built on the LLaDA2.0-Mini diffusion language model backbone. The model is designed for high-quality text-to-image generation and instruction-guided editing.
Rather than relying heavily on paired image-text data from the start, the team first builds a strong visual generative prior through image-only pre-training and mid-training. The generation pipeline comprises 220 million samples, of which 98 are real images. For efficient and scalable optimization, the DiT uses parameter-free RMSNorm throughout, along with the Muon optimizer.
The resulting unified model produces highly photorealistic images while accurately following fine-grained editing instructions. The researchers further distilled LLaDA-Image into LLaDA-Image-Turbo, enabling fast inference in just 2-4 sampling steps.
On the Qwen-Image-Bench benchmark, LLaDA-Image achieved overall scores of 53.53 and 53.38 on the English and Chinese tracks, respectively, setting a new state-of-the-art among open-source models on both tracks. To support further research, the team is releasing model weights, training code, and detailed recipes.