Xie Saining's Team Releases the First Multi-User Video World Model Solaris, Powered by Kunlun Wanzhi Matrix-Game2.0 Foundation

Recently, the team of Xie Saining, an assistant professor at New York University and the author of DiT (Diffusion Transformers), officially released Solaris, the world's first multi-person video world model. This model is based on Matrix-Game 2.0, a general scenario real-time long sequence interactive generation world model that was open-sourced by Kunlun Tian Gong in August 2025, and successfully validated the multi-player collaborative perception capabilities in the game Minecraft.

Solaris achieves efficient information exchange between players by introducing a multi-player self-attention layer. In complex interaction scenarios such as building consistency and player visual alignment, its performance significantly outperforms the previous only multi-player solution, Multiverse.

As the core support of Solaris, the open-sourcing of Matrix-Game 2.0 not only fills the technical gap in interactive generation world models but also demonstrates the key value of domestic open-source foundations in global cutting-edge AI research.

Notably, AMI, an AI company co-founded by Xie Saining as co-founder and Chief Scientific Officer, and Yann LeCun, a Turing Award winner who founded the company in January 2026, recently announced a $1.03 billion seed round financing with a valuation of $3.5 billion.

This technological breakthrough of Solaris and the capital hype of AMI jointly confirm the great potential of world models in the process of moving toward artificial general intelligence (AGI). The continuous evolution and application of Matrix-Game 2.0 will provide important technical foundations for global developers to explore cutting-edge AI applications such as multi-player collaboration and real-time interaction.

Matrix-Game 2.0Open Source Address

Technical Report:

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

Project Homepage:

https://matrix-game-v2.github.io/

HuggingFaceAddress:

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

GitHubAddress:

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

Xie Saining, the author of DiT, releases another breakthrough! Multi-person video world model Solaris launched, with a seed round valuation exceeding $3.5 billion

In March 2026, Xie Saining's team launched Solaris, the world's first multi-person video world model, transitioning from single-view to interactive multi-player environments. Built on the Kunlun Tiangang open-source framework, it uses multi-person self-attention layers to enhance architectural consistency and validate collaborative perception in virtual worlds.....

Alibaba Open Sources Z-Image Image Model: Supports Bilingual Text Rendering

Alibaba open-sources the Z-Image image generation model, which achieves efficient generation and editing with only 6B parameters, reaching visual quality close to 20B commercial models. The model uses a single-stream DiT architecture, offering fast generation and low resource consumption, and has the potential to promote the popularization of AI image tools for consumer applications.

Tencent HunyuanImage 2.1 Makes a Strong Debut! Open-Source 2K Text-to-Image Champion, Turns into High-Resolution Art Master in an Instant?

Recently, the Tencent Hunyuan team officially open-sourced HunyuanImage2.1, this 17B parameter DiT (Diffusion Transformer) text-to-image model quickly topped the Artificial Analysis Image Arena ranking, surpassing HiDream-I1-Dev and Qwen-Image, becoming the new leader in open-source weight models. The model supports native 2048x2048 resolution output, and significantly improves

Kunlun Tech officially launches SkyReels-A3 model: Photos can synchronize with voice

Kunlun Tech Group announced the launch of its latest technological achievement, the SkyReels-A3 model, which is an audio-driven digital human creation tool based on the DiT (Diffusion Transformer) video diffusion model. The release of SkyReels-A3 marks a major advancement in the field of digital content creation. It enables full-modal audio-driven digital human creation for any duration, offering users a brand-new experience. The core function of the SkyReels-A3 model is to bring static images or videos to life.

Kuaishou and Peking University Join Hands to Open Source Video Generation Model Pyramid-Flow: Capable of Generating 10-Second Ultra High Definition Videos

Recently, Kuaishou, along with research teams from Peking University and Beijing University of Posts and Telecommunications, jointly open-sourced an ultra high definition video generation model named Pyramid-Flow. This model can generate videos of up to 10 seconds long, with a resolution of 1280x768 and 24 frames per second, producing high-quality results with excellent lighting effects, action consistency, and overall video quality. Pyramid-Flow operates differently from existing video diffusion models, which typically run at full resolution; while this approach allows for...