~ similar to 2606.11180· 17 results
Xinyin Ma, Julius Berner, Chao Liu, Arash Vahdat +2 more
This paper introduces Flex-Forcing, a framework for video generation that enables a model to operate under both bidirectional and autoregressive generation regimes, achieving better video quality and…
Hongyu Liu, Chun Wang, Feng Gao, Xuanhua He +5 more
This paper proposes OPSD-V, an on-policy self-distillation method for reducing long-horizon degradation in few-step autoregressive video diffusion models by introducing real long-video data as tempora…
Bowen Zheng, Andrew H. Yang, Jiaqi Ruan, Jia He +4 more
The authors present StreamMUSE, an inference system for language models that performs synchronous generation with external signals in real-time.
The paper proposes DLLM-VSR, a novel Diffusion Large Language Model framework for Visual Speech Recognition, achieving state-of-the-art performance by treating transcription as iterative masked denois…
The paper proposes a sequence-alignment framework using Soft Dynamic Time Warping to evaluate audio-driven talking-head generation, demonstrating that this approach provides more robust and fair compa…
Yijie Jin, Jiajun Xu, Yuxuan Liu, Chenkai Xu +7 more
This paper proposes Multi-Block Diffusion Language Models (MBD-LMs) for text generation, which are obtained by post-training Block Diffusion Language Models (BD-LMs) with Multi-block Teacher Forcing (…
This paper introduces Video Diffusion Avatars, a method for creating photo-realistic, multi-view consistent avatars from a single image.
Woojung Han, Seil Kang, Youngjun Jun, Min-Hung Chen +2 more
The paper proposes PhaseLock, a training-free framework that preserves physical consistency in image-to-video diffusion models by extracting a motion prior from few-step inference.
VideoMLA introduces a novel Multi-Head Latent Attention (MLA) mechanism that replaces per-head KV caches with a shared low-rank content latent, significantly reducing memory and improving throughput f…
This paper trains a causal LSTM next-frame predictor on bonafide speech using Wav2Vec2-Large-AntiDeepfake SSL backbone and achieves competitive or state-of-the-art performance on six speech synthesis…
Haoxuan Wu, Lai Man Po, Mengyang Liu, Kun Li +2 more
The paper introduces PRISM, a method for decoding preference signals from noisy latents using a lightweight Query-based Aggregation head and a frozen video diffusion backbone, achieving state-of-the-a…
Haojun Xu, Jie Huang, Xin Lu, Mingchen Zhong +3 more
The paper proposes a new framework, DyMD, to improve video generation and embodied prediction by adapting teacher supervision and critic fitting to the evolving student.
The paper presents LESS, a training-free, model-agnostic adaptive sampler for diffusion large language models that improves accuracy and reduces computational cost.
This paper investigates the phenomenon of 'copying' in Distribution Matching Distillation (DMD), finding that high-dimensional distillation causes student models to spontaneously reproduce the teacher…
This paper conditions a diffusion-based speech enhancement model on wav2vec 2.0 features using Feature-wise Linear Modulation (FiLM), achieving competitive performance on VoiceBank-DEMAND and LibriMix…