Jiaqi Feng

Jiaqi Feng

A step forward

Stream4D: 4D-Consistency for Streaming Autoregressive Diffusion Video Models

Published in arXiv preprint arXiv:2608.19556, 2026

Stream4D improves consistency and preserves natural motion in streaming autoregressive video generation using dynamic 4D reconstruction rewards and a motion prior.

@misc{ban2026stream4d,
  title         = {Stream4D: 4D-Consistency for Streaming Autoregressive Diffusion Video Models},
  author        = {Ban, Yuanhao and Feng, Jiaqi and Zhou, Hengguang and Pei, Xiaohuan and Cui, Justin and Hsieh, Cho-Jui},
  year          = {2026},
  eprint        = {2608.19556},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url           = {https://arxiv.org/abs/2608.19556}
}

Recommended citation: Yuanhao Ban, Jiaqi Feng, Hengguang Zhou, Xiaohuan Pei, Justin Cui, and Cho-Jui Hsieh. "Stream4D: 4D-Consistency for Streaming Autoregressive Diffusion Video Models." arXiv preprint arXiv:2608.19556, 2026.
Download Paper