README.md

metadata

title: StatePlay Game World Model
emoji: 🎮
colorFrom: red
colorTo: pink
sdk: gradio
sdk_version: 6.21.0
app_file: app.py
short_description: Action-conditioned SF3 rollouts with predicted game state
python_version: '3.12'
startup_duration_timeout: 1h
license: apache-2.0
models:
  - onepiece1999/StatePlay
  - Wan-AI/Wan2.2-TI2V-5B

StatePlay — State-Aware Game World Models for Mechanics-Consistent Generation

Interactive demo of StatePlay: a Street Fighter III world model that predicts the game's internal state (round timer, both players' health, both players' super meters) jointly with the video frames, so generated gameplay stays consistent with the underlying game mechanics.

Give it a first frame, an initial game state and a controller sequence (one 0.5 s slot per entry) and it rolls out a 5-second clip at 832×480 / 20 fps plus the predicted state trajectory.

• Paper: https://huggingface.co/papers/2607.26754
• Project page: https://jimntu.github.io/stateplay_page/
• Code: https://github.com/Jimntu/StatePlay
• Checkpoint: https://huggingface.co/onepiece1999/StatePlay (visual expert initialised from Wan-AI/Wan2.2-TI2V-5B)

Street Fighter III visuals are the property of Capcom Co., Ltd. The bundled example frames come from the authors' released held-out evaluation clips and are included for research demonstration only.