Is Yume 1.5 a game engine?

In a way, yes. It is a "neural game engine." Instead of rendering polygons like Unity or Unreal, it "dreams" the next frame of the video based on your inputs, allowing for infinite open world exploration without 3D assets.

Yes, the code and model weights (Yume 5B 720P) are open source and available for free on GitHub and Hugging Face.

Can I run Yume 1.5 on my PC?

It is very demanding. While the code is available, running it at playable speeds currently requires enterprise grade GPUs (like an NVIDIA A100), though "Lite" versions for consumer cards may be developed by the community.

Who created Yume 1.5?

It was developed by researchers at the Shanghai AI Laboratory and Fudan University, led by Xiaofeng Mao and Kaipeng Zhang.

Yume 1.5

A Text Controlled Interactive World Generation Model.

From Shanghai AI Laboratory (Open Source Research)

Founder: Xiaofeng Mao, Kaipeng Zhang, and team (Shanghai AI Laboratory/Fudan University)

Video & Animation

Dec 2025

15K+

Free

Pricing Model

Free

Starting Price

$0/Month

About This AI

Yume 1.5 is an advanced “World Model” that functions like a playable video game generated entirely by AI in real time.

Unlike standard video generators that create a passive movie clip, Yume 1.5 generates an interactive, consistent 3D like environment that users can explore using keyboard controls (WASD) and modify with text commands.

It uses a novel architecture called “Joint Temporal Spatial Channel Modeling” (TSCM) to remember infinite context, ensuring that if you turn around in the virtual world, the scenery remains consistent with what was there before.

3D AI Art Image To Video Text To Video Video Generator

Pricing

Pricing Model

Free

Starting Price

$0/Month

Key Features

Interactive Exploration Allows users to navigate the generated video world in real time using keyboard inputs (WASD for movement), similar to a first person game.
Text Controlled Events Users can type commands to trigger specific events within the world instantly (e.g., "it starts raining," "a cat runs across the street").
Infinite Context Uses TSCM to compress historical frames, allowing for theoretically infinite video generation without the "amnesia" common in older models.
Real Time Speed Optimized with bidirectional attention distillation to run at playable frame rates (approx. 12fps on A100) on single GPUs.
Single Image Start Can build an entire dynamic world starting from just one uploaded image or a simple text prompt.
Dual Stream Attention Separates action processing from visual rendering to prevent control inputs from distorting the image quality.

Pros

The first open source model to combine high quality video generation with game like interactivity.
Solves the "long video consistency" problem better than standard diffusion models.
Completely free and open source (Apache 2.0).
Highly responsive to complex text instructions for environmental changes.

Cons

Requires high end GPU hardware (e.g., A100 or H100) for smooth real time performance.
Resolution is currently limited (often 720p) compared to non interactive renderers.
No user friendly app for non coders (requires Python/CLI setup).
As a research demo, physics interactions are visual only (no true collision detection engine).

Best for AI Researchers, Game Developers, and simulation engineers looking to experiment with “playable” generative video and neural game engines.

FAQs

Is Yume 1.5 a game engine?
In a way, yes. It is a “neural game engine.” Instead of rendering polygons like Unity or Unreal, it “dreams” the next frame of the video based on your inputs, allowing for infinite open world exploration without 3D assets.
Is Yume 1.5 free?
Yes, the code and model weights (Yume 5B 720P) are open source and available for free on GitHub and Hugging Face.
Can I run Yume 1.5 on my PC?
It is very demanding. While the code is available, running it at playable speeds currently requires enterprise grade GPUs (like an NVIDIA A100), though “Lite” versions for consumer cards may be developed by the community.
Who created Yume 1.5?
It was developed by researchers at the Shanghai AI Laboratory and Fudan University, led by Xiaofeng Mao and Kaipeng Zhang.