WHAM 1.6B (Muse)
Release in the Muse (Microsoft Research WHAM) family · version WHAM 1.6B v1
Maintained by Microsoft Research1
WHAM 1.6B is the larger of the two World and Human Action Model checkpoints Microsoft Research released for Muse. It pairs a VQ-GAN image tokenizer with a transformer trained from scratch to predict game visuals and controller actions. The same repository includes a 200M-parameter checkpoint for lighter experiments.1
- Model hub: Model card (Hugging Face) (external site: huggingface.co)
- License: License (Microsoft Research License Terms) (external site: huggingface.co)
- Dataset hub: Sample gameplay data (external site: huggingface.co)
- Paper: Nature paper (external site: nature.com)
Availability and license
Overall availability
Checkpoints can be downloaded from Hugging Face without a gated access request, but the license allows only non-commercial research use and forbids redistributing the models. Microsoft released the weights in February 2025 along with the WHAM Demonstrator and sample data.324
Availability is separate from permission: read the license before using or redistributing.
The Microsoft Research License covers the source code, object code, models, and data in the release. It allows use only for non-commercial, non-revenue-generating research. Licensees may modify the source code but may not distribute the code, models, or data. They may not host the materials as a stand-alone service for others or reverse engineer them. Published results may not include substantial portions of the materials. U.S. users are bound to individual arbitration. The model card asks users not to remove watermark and provenance metadata from generated images.21
Model-disclosure tier
The model parameters for this release can be downloaded by the public. License terms may still restrict use, redistribution, or commercial use.
The weights are under a license that is not on the rubric's OSI-approved list. Read its terms before use.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| WeightsCan the general public download the model parameters for this release? | Public | WHAM 1.6B and 200M checkpoints are published without gating, under a research-only, non-commercial license.312 |
| Inference codeIs code for running the model published? | Public | The repository includes setup scripts and a run_dreaming.py script for local inference, plus the WHAM Demonstrator application.1 |
| Training codeIs the code used to train the model published? | Unknown | Not assessed. |
| Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access. | Partial | The model card describes about 500,000 Bleeding Edge games (more than 1 billion observation-action pairs at 10 Hz) from anonymized gameplay by 27,990 players. Only a sample, the evaluation test data, is published.1 |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Unknown | Not assessed. |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Partial | The model card describes the evaluation measures (FVD for consistency, Wasserstein distance for diversity, and persistency tests) and publishes the test data sample. Detailed results are in the Nature paper.1 |
What it is useful for
The model card describes academic research use, such as generating plausible gameplay sequences for the game Bleeding Edge in world-modelling, behavior-policy, or combined modes. It is not intended to generate imagery outside that game.1
Run and use notes
- The model card documents a setup script that creates a Python 3.9 environment. It says testing was done on Ubuntu 20.04 (natively and under WSL2 on Windows 11) with NVIDIA RTX A6000 and A100 GPUs, and that a CUDA-capable GPU is assumed for inference.1
Organization context
Provenance and derivatives
The model card says both components were trained from scratch by Microsoft Research on gameplay data from Ninja Theory's Bleeding Edge.1
Other releases in the Muse (Microsoft Research WHAM) family
No other releases in this family have been assessed.
U.S. eligibility
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.