Matrix-Game 2.0 Released: The First Open-Source Interactive World Model for Real-Time Long-Sequence Generation

On August 12, Skywork AI announced the release of Matrix-Game 2.0, the upgraded version of its Matrix series interactive world model. This breakthrough model delivers real-time, long-sequence interactive video generation across general-purpose scenarios, and the model is fully open-sourced, making it the first of its kind in the industry.

Matrix-Game 2.0 represents a major leap in both real-time performance and long-sequence generation capabilities. With a focus on low latency and high frame rates, the model can stably generate continuous video at 25 FPS across complex environments, with durations extending to minutes. The result is significantly enhanced coherence, usability, and immersion.

In addition to faster inference, Matrix-Game 2.0 maintains precise understanding of physics and scene semantics. Users can issue simple commands to freely explore, manipulate, and construct virtual environments that are structurally consistent, visually rich, and logically sound in real time. This breakthrough removes the barrier between content generation and interactive engagement, opening new possibilities for applications in virtual humans, gaming engines, embodied AI, and more.

Model Architecture Matrix-Game 2.0 introduces a new vision-driven approach to interactive world modeling-moving away from language-prompt dependency and focusing on spatial understanding and physics-based learning.

Image: https://lh7-rt.googleusercontent.com/docsz/AD_4nXdeiWngcYGRyK9byjFAqgK40v2Vs73DgIBCXOEKjvWeqbTP4HBpz9DixAnuBWPUvczzcqwWrSGMPMk_tk20ilZqDpjH63EoGqk-ylZn4wx6bt40HzNeHThSc3pbo-tiOu22NGUI?key=y1A95UT3dg9dvOAiAmmbvA

*
3D Causal VAE Compression: Efficiently compresses spatial and temporal dimensions for better modeling and generation.

*
Multimodal Diffusion Transformer (DiT): Combines vision encoding with user action commands to generate frame-by-frame realistic dynamic sequences.

*
User Interaction Module: Adapts GameFactory and Genie-style frameworks to enable real-time control.

Image: https://lh7-rt.googleusercontent.com/docsz/AD_4nXeonwkuQGyiYUko7vKPUHspDgvPA1eO0aZgNB4WDu9g2sSLVNLBLoCOAup2oeoWZejV3txbXQzxsIbFYGtv1gPjC0OqUkRFYGuyThE22rLWO-uawx65U_kN9lagzQ-xF6GMK2bdzA?key=y1A95UT3dg9dvOAiAmmbvA

Real-Time Autoregressive Video Generation Using a Self-Forcing training strategy, Matrix-Game 2.0 employs a novel autoregressive diffusion generation mechanism to overcome latency and error accumulation in conventional models:

*
Causal Diffusion Model Distillation: Minimizes sequence delay by conditioning on past frames.

*
Distribution Matching Distillation (DMD): Aligns training and inference distributions for more stable results.

*
KV Cache Mechanism: Enables seamless long video generation without redundant computation, supporting unlimited output length at 25 FPS on a single GPU.

Applications & Performance Matrix-Game 2.0 supports dynamic, physics-consistent interactions-such as character movement and camera rotation-through keyboard and mouse input. It is applicable to diverse scenes, including GTA-style environments, Minecraft, and open-world exploration, with enhanced cross-domain adaptability and physical realism.

Three Core Breakthroughs:

*
High-FPS Real-Time Long-Sequence Generation: Minute-long, natural, and responsive interactions at 25 FPS.

*
Multi-Scene Generalization: Adaptable to various styles and environments, from urban landscapes to artistical renderings.

*
Enhanced Physical Consistency: Realistic movement over complex terrains, boosting immersion and controllability.

Matrix-Game 2.0 sets a new milestone for spatial intelligence research and application, paving the way for embodied AI training, rapid virtual world construction, and content creation for films and the metaverse.

Image: https://lh7-rt.googleusercontent.com/docsz/AD_4nXdhI460P2BZuzzDVCC7Ex9e6oePKYnLQ-89VYA5_jT03VyhskYwYR_ZLAPJOBmWI2o4hE_YBveoNtksqT8Y-PXP-6nIppBCEeXG4ZeRixtze_xmA5nqbqusUWoGE_4KYJ-skCJa?key=y1A95UT3dg9dvOAiAmmbvA

Open-Source Links:

*
Technical Report: https://github.com/SkyworkAI/Matrix-Game/blob/main/Matrix-Game-2/assets/pdf/report.pdf

*
Project Page:https://matrix-game-v2.github.io/

*
HuggingFace: https://huggingface.co/Skywork/Matrix-Game-2.0

*
GitHub: https://github.com/SkyworkAI/Matrix-Game

Skywork.ai [http://skywork.ai] is a consumer-facing AI workspace and creative platform that helps everyday users produce slides, spreadsheets, videos, documents, and interactive content in minutes - built around intuitive conversational workflows. The platform offers guided prompts, real-time previews, and integrations with common office tools to speed up workflows for students, freelancers, and small teams. Available on web and mobile, http://skywork.ai emphasizes ease of use, affordability, and rapid iteration-bringing advanced AI creativity tools directly to consumers.
Media Contact
Company Name: Skywork AI PTE.LTD.
Contact Person: Peter Tian
Email: Send Email [http://www.universalpressrelease.com/?pr=matrixgame-20-released-the-first-opensource-interactive-world-model-for-realtime-longsequence-generation]
Country: Singapore
Website: https://skywork.ai/

Legal Disclaimer: Information contained on this page is provided by an independent third-party content provider. GetNews makes no warranties or responsibility or liability for the accuracy, content, images, videos, licenses, completeness, legality, or reliability of the information contained in this article. If you are affiliated with this article or have any complaints or copyright issues related to this article and would like it to be removed, please contact retract@swscontact.com



This release was published on openPR.