arXiv:2505.23606v5 Announce Type: replace 
Abstract: Unified generation models aim to handle diverse tasks across modalities -- such as text generation, image generation, and vision-language reasoning -- within a single architecture and decoding paradigm. Autoregressive unified models suffer from slow inference due to sequential decoding, and non-autoregressive unified models suffer from weak generalization due to limited pretrained backbones. We introduce the second-generation Meissonic: Muddit, a unified discrete diffusion transformer that enables fast and parallel generation across both text and image modalities. Unlike prior unified diffusion models trained from scratch, Muddit integrates strong visual priors from a pretrained text-to-image backbone with a lightweight text decoder, enabling flexible and high-quality multimodal generation under a unified architecture. Empirical results show that Muddit achieves competitive or superior performance compared to significantly larger autoregressive models in both quality and efficiency. The work highlights the potential of purely discrete diffusion, when equipped with strong visual priors, as a scalable and effective backbone for unified generation.

يمثل تقديم Muddit، وهو نموذج تحويلي موحد من الجيل الثاني، تقدماً كبيراً في مجال التوليد متعدد الوسائط، حيث يتيح إنشاء النصوص والصور بسرعة وبشكل متوازي. يدمج هذا النموذج أولويات بصرية قوية من قاعدة بيانات مسبقة التدريب لتحويل النص إلى صورة، مما يعزز جودة ومرونة النتائج مقارنة بالنماذج السابقة.

La introducción de Muddit, un transformador de difusión discreto unificado de segunda generación, representa un avance significativo en la generación multimodal, permitiendo la creación rápida y paralela de texto e imágenes. Este modelo integra fuertes prioridades visuales de una base preentrenada de texto a imagen, mejorando la calidad y flexibilidad de los resultados en comparación con modelos anteriores.

L'introduction de Muddit, un transformateur de diffusion discret unifié de deuxième génération, représente une avancée significative dans la génération multimodale, permettant la création rapide et parallèle de textes et d'images. Ce modèle intègre de forts prioris visuels d'une base pré-entraînée de texte à image, améliorant la qualité et la flexibilité des résultats par rapport aux modèles précédents.

The introduction of Muddit, a second-generation unified discrete diffusion transformer, represents a significant advancement in multimodal generation, enabling fast and parallel creation of both text and images. This model integrates strong visual priors from a pretrained text-to-image backbone, enhancing the quality and flexibility of outputs compared to previous models.