✦
Summary
Code World Model uses a coding agent to update and control the state of a generated world. It separates persistent world structure from visual appearance, allowing the environment to continue evolving while its rendering style changes, with applications in interactive world generation and controllable synthetic experiences.
Key Features
The agent updates and controls the evolving world state.
The demonstrations show persistent state across periodic visual-style changes.
Examples compare generated RGB output with an underlying proxy representation.
The public inference stack includes a MiniMax-H3-based CWM LoRA.
The repository supplies forty compact examples with released configurations.
Compact NPZ conditions are expanded locally for inference.
The released command-line workflow includes configuration validation and preparation.
Separate environment, inference, license, and third-party notice files support setup.
About Code World Model
The approach gives the agent responsibility for world-state reasoning and modification, then connects that state to a video-generation pipeline. The released inference stack includes a MiniMax-H3-based LoRA and compact example conditions. Explicit state supports continuity across longer visual sequences and repeated stylistic transformations.
The project is useful for researchers exploring a programmable layer beneath video world models. Its public repository supplies environment guidance, validated example configurations, and generation commands. Running the examples requires downloading the associated assets and using the documented model environment, including applicable third-party model terms.