VideoMAR: Autoregressive Video Generatio with Continuous Tokens

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Yu, Hu, Gong, Biao, Yuan, Hangjie, Zheng, DanDan, Chai, Weilong, Chen, Jingdong, Zheng, Kecheng, Zhao, Feng
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908412501032960
author Yu, Hu
Gong, Biao
Yuan, Hangjie
Zheng, DanDan
Chai, Weilong
Chen, Jingdong
Zheng, Kecheng
Zhao, Feng
author_facet Yu, Hu
Gong, Biao
Yuan, Hangjie
Zheng, DanDan
Chai, Weilong
Chen, Jingdong
Zheng, Kecheng
Zhao, Feng
contents Masked-based autoregressive models have demonstrated promising image generation capability in continuous space. However, their potential for video generation remains under-explored. In this paper, we propose \textbf{VideoMAR}, a concise and efficient decoder-only autoregressive image-to-video model with continuous tokens, composing temporal frame-by-frame and spatial masked generation. We first identify temporal causality and spatial bi-directionality as the first principle of video AR models, and propose the next-frame diffusion loss for the integration of mask and video generation. Besides, the huge cost and difficulty of long sequence autoregressive modeling is a basic but crucial issue. To this end, we propose the temporal short-to-long curriculum learning and spatial progressive resolution training, and employ progressive temperature strategy at inference time to mitigate the accumulation error. Furthermore, VideoMAR replicates several unique capacities of language models to video generation. It inherently bears high efficiency due to simultaneous temporal-wise KV cache and spatial-wise parallel generation, and presents the capacity of spatial and temporal extrapolation via 3D rotary embeddings. On the VBench-I2V benchmark, VideoMAR surpasses the previous state-of-the-art (Cosmos I2V) while requiring significantly fewer parameters ($9.3\%$), training data ($0.5\%$), and GPU resources ($0.2\%$).
format Preprint
id arxiv_https___arxiv_org_abs_2506_14168
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle VideoMAR: Autoregressive Video Generatio with Continuous Tokens
Yu, Hu
Gong, Biao
Yuan, Hangjie
Zheng, DanDan
Chai, Weilong
Chen, Jingdong
Zheng, Kecheng
Zhao, Feng
Computer Vision and Pattern Recognition
Artificial Intelligence
Masked-based autoregressive models have demonstrated promising image generation capability in continuous space. However, their potential for video generation remains under-explored. In this paper, we propose \textbf{VideoMAR}, a concise and efficient decoder-only autoregressive image-to-video model with continuous tokens, composing temporal frame-by-frame and spatial masked generation. We first identify temporal causality and spatial bi-directionality as the first principle of video AR models, and propose the next-frame diffusion loss for the integration of mask and video generation. Besides, the huge cost and difficulty of long sequence autoregressive modeling is a basic but crucial issue. To this end, we propose the temporal short-to-long curriculum learning and spatial progressive resolution training, and employ progressive temperature strategy at inference time to mitigate the accumulation error. Furthermore, VideoMAR replicates several unique capacities of language models to video generation. It inherently bears high efficiency due to simultaneous temporal-wise KV cache and spatial-wise parallel generation, and presents the capacity of spatial and temporal extrapolation via 3D rotary embeddings. On the VBench-I2V benchmark, VideoMAR surpasses the previous state-of-the-art (Cosmos I2V) while requiring significantly fewer parameters ($9.3\%$), training data ($0.5\%$), and GPU resources ($0.2\%$).
title VideoMAR: Autoregressive Video Generatio with Continuous Tokens
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2506.14168