gracezhao1997/Awesome-Video-World-Models-with-AR-Diffusion

A Curated List of Awesome Video World Models with AR Diffusion: Covering Algorithms, Applications, and Infrastructure, Aimed at Serving as a Comprehensive Resource for Researchers, Practitioners, and Enthusiasts.

What this is

This is a curated list of research papers and resources about a specific type of AI video model: video world models built with autoregressive diffusion (AR diffusion). These models generate videos chunk-by-chunk (autoregressive) while each chunk is produced using diffusion (a process that gradually turns noise into images or video). The repository aims to be a comprehensive, structured guide for researchers, developers, and enthusiasts in this field.

What it covers

The list is organized into three main dimensions:

Algorithmic Foundations – methods for training such models, making them run faster, and generating longer videos.
Real-world Applications – open-source video foundation models, interactive action-conditioned models, avatars, egocentric interactions, and use in embodied AI/autonomous driving.
Infrastructure-level Acceleration – techniques like sparse attention, caching, quantized attention, and speculative decoding that make these models run efficiently.

Key features

Structured taxonomy: papers are grouped into clear categories (see the table of contents) to help you navigate many approaches.
One-stop citation collection: a single BibTeX file (video-world-models.bib) lists all papers, so you can import them into LaTeX or Zotero with one click.
Weekly updates: the maintainers refresh the list every week to keep it current.
Community contributions: the maintainers explicitly welcome pull requests to add missing papers, and also accept high-level contributions like better categorizations or syntheses.

New highlight: minWM

A recent addition is minWM – a full-stack open-source framework that shows how to turn a bidirectional text-to-video model into an action-conditioned video world model, with example data, scripts, and documentation. It also has a technical report and pre-trained models on Hugging Face.

Maintenance and contact

The repository is maintained by Min Zhao, Hongzhou Zhu, Wenqiang Sun, and Bokai Yan. The README provides their contacts and asks you to reach out if you have questions or suggestions.

How to use this repo

As a research guide: browse the categories to discover key methods in AR diffusion for video world modeling.
As a citation tool: grab the BibTeX file for your papers.
As a learning resource: the included papers and links to code/websites give you direct access to implementations and project pages.