VideoWorldModel | CVPR 2026 Workshop

About

World models are systems that predict the next state of the world based on historical states and interactive action control. Among various approaches, Video World Models based on interactive video generation are particularly promising due to the photorealism and scalability of video data, as well as recent advances in video generation.

However, current video generation models face significant challenges in achieving the ideal Video World Model. This workshop aims to provide a platform for researchers from both academia and industry to discuss and address these challenges, foster collaboration, advance related academic research, and promote the practical application of Video World Models.

🎮

Interaction

Effective interaction with virtual worlds, including navigation and object manipulation

🧠

Memory

Maintaining consistency over long video sequences with causal reasoning

⚡

Efficiency

Real-time video generation with high quality, addressing throughput and latency

🤖

Applications

Robotics, Embodied AI, autonomous driving, and more

Schedule

Workshop Date: June 3rd

Time Slot: AM session, 8:00 AM - 1:00 PM

Room: Room 705/707

Poster Location: Posters for all workshops will be presented in Exhibit Hall A.

Event	Speaker	Time
Opening Remarks		8:30 AM - 8:35 AM
Invited Talk #1 (20 min + 5 min Q&A)	Jack Parker-Holder	8:35 AM - 9:00 AM
Invited Talk #2 (20 min + 5 min Q&A)	Andrea Vedaldi	9:00 AM - 9:25 AM
Invited Talk #3 (20 min + 5 min Q&A)	Sherry Yang	9:25 AM - 9:50 AM
Poster Session + Coffee Break + Best Paper Award		9:50 AM - 10:40 AM
Invited Talk #4 (20 min + 5 min Q&A)	Yilun Du	10:40 AM - 11:05 AM
Invited Talk #5 (20 min + 5 min Q&A)	Xingang Pan	11:05 AM - 11:30 AM
Invited Talk #6 (20 min + 5 min Q&A)	Yaoyao Liu	11:30 AM - 11:55 AM
Closing Remarks		11:55 AM - 12:00 PM

Call for Papers

Topics of Interest

Interactive video generation and world simulation
Long-term memory and consistency in video generation
Efficient and real-time video generation methods
Video world models for robotics and embodied AI
Video world models for autonomous driving
Action-conditioned video prediction and generation
Evaluation and benchmarking of video world models
Novel architectures and training methods for video world models

Track 1: Proceedings

Submissions must present original, unpublished research. Manuscripts should be 4–8 pages (excluding references) using the CVPR 2026 template. Accepted papers will be published in the CVPR 2026 Workshop Proceedings.

Submission Deadline March 1, 2026 (AoE)
Notification to Authors March 20, 2026
Camera-Ready Deadline April 10, 2026

Submission Portal

Track 2: Non-Proceedings

A flexible, non-archival venue for sharing a broad range of contributions without restrictive publishing constraints, formatting requirements, or page limits. We warmly welcome:

Works-in-progress and preliminary results
Open Datasets
Technical Reports
Recent work submitted or published within the last year
Position papers and conceptual frameworks

Submission Deadline April 14, 2026 (AoE)
Notification to Authors May 4, 2026
Camera-Ready Deadline May 21, 2026

Submission Portal

Submission Guidelines

Proceedings Track papers must follow the official CVPR 2026 template and be anonymized for double-blind review
Supplementary materials (videos, code, etc.) are welcome

Accepted Papers

We are pleased to feature accepted work across both proceedings and non-proceedings tracks.

3 Proceedings Track Papers Archival workshop proceedings

25 Non-Proceedings Track Papers Non-archival workshop posters

👑 Best Paper Award NeoVerse: Enhancing 4D World Model with in-the-wild Monocular Videos

Proceedings Track

3 papers

GameCraft-2: Instruction-following Interactive Game World Model
Causal Physics Steering in Video World Models via Concept Activation Vectors
CoDA: Cosmos-driven DCT-Adaptive Sparse Attention for Efficient Robotic Trajectory Generation

Non-Proceedings Track

25 papers

Inference-Time Planning with Action-Conditioned Video Models for Generalizable Robot Manipulation
WorldPack: Dynamic Frame Compression for Long-context Video World Modeling
Future Optical Flow Prediction Improves Robot Control & Video Generation
OmniView: An All-Seeing Diffusion Model for 3D and 4D View Synthesis
SEGAR: Selective Enhancement for Generative Augmented Reality
The Pulse of Motion: Measuring Physical Frame Rate from Visual Dynamics
Forecasting Animal Motion in the Wild
NeoVerse: Enhancing 4D World Model with in-the-wild Monocular Videos 👑 Best Paper Award
Objects in Generated Videos Are Slower Than They Appear: Models Suffer Sub-Earth Gravity and Don't Know Galileo's Principle...for now
EgoControl: Controllable Egocentric Video Generation via 3D Full-Body Poses
Dexterous World Models
Causal State Entropy Bounds on Predictive Horizons in Video World Models
Causal State Compression for Long-Horizon Video World Models: A Bounded-Drift Theory and Efficient Architecture
Spectral World Models: Provably Consistent Long-Horizon Video Generation via Koopman Operator Decomposition
Building a Precise Video Language with Human–AI Oversight
Olaf-World: Orienting Latent Actions for Video World Modeling
DECOMWM: Interpretable Reward Decomposition for World-Model-Based Trajectory Selection
WorldCam: Interactive Autoregressive 3D Gaming Worlds with Camera Pose as a Unifying Geometric Representation
RoboWM-Bench: A Benchmark for Evaluating World Models in Robotic Manipulation
When Predictive World Models Meet Diffusion
Is Your Driving World Model an All-Around Player?
WFM-Eval: Interpretable Error Diagnostics for Video World Models in Robotics
VerseCrafter: Dynamic Realistic Video World Model with 4D Geometric Control
Rays as Pixels: Learning a Joint Distribution of Videos and Camera Trajectories
Video Models Reason Early: Exploiting Plan Commitment for Maze Solving

WorldArena Challenge

⚡ Open Now — Join the Challenge

An associated challenge of the Workshop on Video World Models, built on the WorldArena benchmark. The challenge evaluates embodied world models from two complementary perspectives: Track 1: Video Perception Quality and Track 2: Embodied Task Functionality. Submit your model, climb the leaderboard, and compete for awards at CVPR 2026.

Final Submission Deadline in

--Days

--Hours

--Mins

--Secs