7 papers
Back on Track: Aligning Rewards and States for Reasoning in Diffusion Large Language Models
Yawen Shao, Jie Xiao, Kai Zhu +6
Reinforcement learning (RL) holds immense promise for enhancing the reasoning capabilities of diffusion large language models (dLLMs). However, progress is fundamentally constraine…
Wan-Image: Pushing the Boundaries of Generative Visual Intelligence
Chaojie Mao, Chen-Wei Xie, Chongyang Zhong +55
We present Wan-Image, a unified visual generation system explicitly engineered to paradigm-shift image generation models from casual synthesizers into professional-grade productivi…
Towards Sequence Modeling Alignment between Tokenizer and Autoregressive Model
Pingyu Wu, Kai Zhu, Yu Liu +6
Autoregressive image generation aims to predict the next token based on previous ones. However, this process is challenged by the bidirectional dependencies inherent in conventiona…
FACM: Flow-Anchored Consistency Models
Yansong Peng, Kai Zhu, Yu Liu +4
Continuous-time Consistency Models (CMs) promise efficient few-step generation but face significant challenges with training instability. We argue this instability stems from a fun…
Anchoring Values in Temporal and Group Dimensions for Flow Matching Model Alignment
Yawen Shao, Jie Xiao, Kai Zhu +4
Group Relative Policy Optimization (GRPO) has proven highly effective in enhancing the alignment capabilities of Large Language Models (LLMs). However, current adaptations of GRPO…
VanGogh: A Unified Multimodal Diffusion-based Framework for Video Colorization
Zixun Fang, Zhiheng Liu, Kai Zhu +5
Video colorization aims to transform grayscale videos into vivid color representations while maintaining temporal consistency and structural integrity. Existing video colorization…