AI

Singapore, Aug. 11, 2025 (GLOBE NEWSWIRE) — 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.

• 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.

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:

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

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

3. 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.

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 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, skywork.ai emphasizes ease of use, affordability, and rapid iteration—bringing advanced AI creativity tools directly to consumers.

CONTACT: Peter Tian
Skywork AI PTE.LTD.
peter (at) skywork.ai

DUBAI, United Arab Emirates, Aug. 04, 2025 (GLOBE NEWSWIRE) — Micropolis Holding Co. (“Micropolis” or the “Company”) (NYSE: MCRP), a pioneer in unmanned ground vehicles and AI-driven security solutions, today announced it has commenced the final phase of its flagship pilot project in collaboration with Dubai Expo City, Dubai Police, and Transguard Group.

This milestone represents a significant advancement in the evolution of Micropolis’s M2 Autonomous Police Patrol, which will undergo extensive testing and validation across its full operational suite. The project will focus on real-world performance evaluation of the enhanced autonomous driving system, the mission planner, tele-operator control functions, and the newly developed radio communication protocol.

In parallel, Micropolis and its partners will conduct rigorous testing of the integrated AI models designed for facial recognition, suspect tracking, and behavior analysis, which are critical capabilities for enhancing urban safety and law enforcement efficiency.

As part of its proprietary technology suite, Micropolis will also deploy and validate its in-house developed edge computing units, built on the AJX NVIDIA Orin platform. These advanced units will be installed on the autonomous patrol vehicles and strategically positioned CCTV cameras, providing enhanced on-site processing capabilities and enabling faster decision-making at the network edge.

The final testing phase is scheduled to commence in mid-August 2025 and continue through the end of September 2025, establishing the foundation for broader deployment of the M2 platform in public safety and smart city initiatives.

“This final phase of the pilot represents a pivotal moment for autonomous policing technology,” said Fareed Aljawhari, Founder & CEO of Micropolis. “Years of rigorous development and strategic partnerships have brought us to this critical juncture where we can demonstrate the M2 platform’s transformative potential for urban security and public safety operations.”

This project reinforces Micropolis’s commitment to pioneering advanced robotics and AI technologies that are tailored to meet the complex demands of modern urban environments.

About Micropolis Holding Co.
Micropolis is a UAE-based company specializing in the design, development, and manufacturing of unmanned ground vehicles (UGVs), AI systems, and smart infrastructure for urban, security, and industrial applications. The Company’s vertically integrated capabilities cover everything from mechatronics and embedded systems to AI software and high-level autonomy.

For more information please visit www.micropolis.ai.

Forward-Looking Statements
This press release contains “forward-looking statements” within the meaning of the “safe harbor” provisions of the Private Securities Litigation Reform Act of 1995You can identify forward-looking statements by the fact that they do not relate strictly to historical or current facts. These statements may include words such as “anticipate”, “estimate”, “expect”, “project”, “plan”, “intend”, “believe”, “may”, “will”, “should”, “can have”, “likely” and other words and terms of similar meaning. Forward-looking statements represent Micropolis’ current expectations regarding future events and are subject to known and unknown risks and uncertainties that could cause actual results to differ materially from those implied by the forward-looking statements. These statements are subject to uncertainties and risks including, but not limited to, the uncertainties related to market conditions and other factors discussed in the “Risk Factors” section of the registration statement filed by the Company with the SEC. For these reasons, among others, investors are cautioned not to place undue reliance upon any forward-looking statements in this press release. Additional factors are discussed in the Company’s filings with the SEC, which are available for review at www.sec.gov. The Company undertakes no obligation to publicly revise these forward-looking statements to reflect events or circumstances that arise after the date hereof.

Investor Contact:
KCSA Strategic Communications
Valter Pinto, Managing Director
PH: (212) 896-1254
Valter@KCSA.com

Media Contact:
Jessica Starman
media@elev8newmedia.com

A photo accompanying this announcement is available at https://www.globenewswire.com/NewsRoom/AttachmentNg/9be0672d-b58a-45b0-a07e-d826103157ac