AI RESEARCH

Steering Video Diffusion Transformers with Massive Activations

arXiv CS.CV

ArXi:2603.17825v1 Announce Type: new Despite rapid progress in video diffusion transformers, how their internal model signals can be leveraged with minimal overhead to enhance video generation quality remains underexplored. In this work, we study the role of Massive Activations (MAs), which are rare, high-magnitude hidden state spikes in video diffusion transformers.