Trends · 研究趋势
数据来自 Hugging Face 论文的 AI 提取关键词,按月统计研究方向的增长与热度。
WanTeam, Ang Wang, Baole Ai +59 authors
Wan, a comprehensive suite of video foundation models built on the diffusion transformer paradigm, advannces video generation by introducing a novel VAE, scalable pre-training strategies, and large-scale data curation, offering superior performance and versatility across various applications with both large and efficient models.
Jintao Zhang, Chendong Xiang, Haofeng Huang +4 authors
SpargeAttn is a universal sparse and quantized attention method that accelerates diverse models by predicting the attention map and using an online softmax-aware filter to skip unnecessary matrix multiplications.
Jintao Zhang, Haofeng Huang, Pengle Zhang +3 authors
SageAttention2 improves attention computation speed with 4-bit matrix multiplication and adaptive quantization, achieving significant OPS increases without compromising accuracy.
Yu Lu, Yuanzhi Liang, Linchao Zhu +1 authors
FreeLong enhances long video generation by balancing global and local video features during denoising, improving quality and consistency.
Jintao Zhang, Jia wei, Pengle Zhang +2 authors
A new quantization method for attention mechanisms, named SageAttention, enhances inference speed and accuracy without significant loss in performance across various models.
Donghyun Lee, Jitesh Chavan, Duy Nguyen +5 authors
OrbitQuant enables efficient post-training quantization for diffusion transformers by using a normalized rotated basis that eliminates the need for recalibration across different timesteps and modalities.
Shenghai Yuan, Jinfa Huang, Xianyi He +5 authors
ConsisID is a tuning-free, DiT-based IPT2V model that uses frequency-aware identity control to generate high-quality, identity-preserving videos by decomposing facial features into low and high-frequency elements.
Runjia Qian, Zile Wang, Jihai Zhang +14 authors
Matrix-Game 3.5 improves interactive world generation via geometry-aware memory, static-dynamic disentanglement, and progressive distillation for long-horizon real-time simulation.
Yabo Zhang, Yuxiang Wei, Dongsheng Jiang +3 authors
ControlVideo, a training-free framework adapted from ControlNet, improves text-to-video generation by ensuring appearance coherence, reducing flicker, and efficiently producing long videos.
北京市昌平区探索星信息技术及软件开发工作室
京ICP备2026059466号