Matrix-Game 3.5 is a memory-augmented interactive world model for 720p long-horizon camera-controllable video generation in first- and third-person modes, featuring a novel Patch Memory and real-time distillation.

Stars

118

7-day growth

No data

Forks

3

Open issues

1

License

Apache-2.0

Last updated

2026-07-27

AI repository intelligence
FR-AI / ANALYSIS

Why it is worth attention

It introduces a parameter-free geometric memory framework (Patch Memory + Warped PRoPE) and a static-dynamic decoupled world representation, enabling real-time, few-step interactive generation with long-horizon consistency and precise camera control.

Who it is for

  • Researchers in video generation and world models
  • Game developers seeking interactive AI-driven environments
  • AI engineers working on real-time simulation and robotics
  • Academics studying long-horizon consistency and autoregressive distillation

Use cases

  • Interactive virtual environment exploration with camera control
  • Real-time video generation for gaming or simulation
  • Egocentric (first-person) video generation for embodied AI
  • Controllable scene generation with subject identity locking

Strengths

  • Parameter-free Patch Memory achieves long-horizon geometric consistency without backbone modification
  • Real-time distillation enables few-step (3-step) causal generation from a bidirectional model
  • Static-dynamic decoupling unifies geometric and subject coherence, reducing ghosting and identity drift
  • Supports both first-person and third-person modes with optional protagonist identity control

Considerations

  • Requires a GPU with at least 40 GB VRAM (A/H series tested), limiting accessibility
  • Only Linux is officially supported with Python 3.10
  • Built on Wan2.2-TI2V-5B backbone; dependency on external depth and text encoders increases deployment complexity

README quick start

Installation

Description

Matrix-Game 3.5: Enhancing Real-Time Streaming Interactive World Models with Patch Memory

Related repositories

Similar projects matched by category, topics, and programming language.

lopopolo
Featured
lopopolo GitHub avatar

harness-engineering

Harness Engineering is a methodology for improving coding agent outputs by carefully crafting the environment around them—providing curated context, tools, and executable constraints that encode an organization’s nonfunctional requirements and cumulative lessons.

AI & Machine LearningAI Agents
2,390
slvDev
Featured
slvDev GitHub avatar

esp32-ai

A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.

AI & Machine LearningLarge Language Models
1,960
littledivy
Featured
littledivy GitHub avatar

mimic

mimic captures traffic from any iOS or web app and automatically generates a Python client library that lets you call the app's API like a regular library.

AI & Machine Learning
1,482