Repolode
Submit
Tencent-Hunyuan

Tencent-Hunyuan/Hunyuan-GameCraft-1.0

Open-source interactive game video generation model that turns a reference image and keyboard or mouse inputs into controllable gameplay-style video.

View on GitHubWebsite
Stars
760
Forks
78
Open issues
19
Last commit
11 months ago

What is it?

Hunyuan-GameCraft is a video generation model from Tencent Hunyuan aimed at interactive game footage. Given a reference image, a text prompt, and a sequence of keyboard or mouse actions such as W, A, S, D, the model produces a video that follows those camera movements. Keyboard and mouse inputs are mapped into a shared camera representation space so that movements can be interpolated smoothly.

The project targets long video extension. It uses a hybrid history-conditioned training strategy that extends sequences autoregressively while keeping scene information from earlier frames. A distilled variant is also provided, which uses 8 inference steps instead of 50 and supports FP8 for faster generation.

The repository includes inference scripts for multi-GPU parallel generation on Linux with CUDA, a low-VRAM single-GPU mode with CPU offload, and a Gradio web interface backed by a distributed inference API. Model weights are published on Hugging Face.

Key features

  • 01Maps keyboard and mouse actions into a shared camera space for fine-grained control.
  • 02Extends videos autoregressively using hybrid history conditioning.
  • 03Provides a distilled model that needs only 8 inference steps.
  • 04Supports FP8 optimization and SageAttention.
  • 05Offers a low-VRAM single-GPU mode with CPU offload.
  • 06Includes a Gradio web interface with a distributed inference API.

Use cases

  • Researching controllable, action-conditioned video generation for games.
  • Prototyping game-style clips from a single concept image and movement inputs.
  • Experimenting with long-horizon video extension on multi-GPU Linux servers.

Quick start

  1. 1

    Clone the repository

    git clone https://github.com/Tencent-Hunyuan/Hunyuan-GameCraft-1.0.git
  2. 2

    Create the conda environment

    conda create -n HYGameCraft python==3.10
  3. 3

    Install PyTorch and pip dependencies

    python -m pip install -r requirements.txt
  4. 4

    Download the pretrained model weights (see weights/README.md)

  5. 5

    Run the Gradio web interface

    bash scripts/launch_app_sp.sh

Things to know

  • Requires an NVIDIA GPU with CUDA; at least 24GB GPU memory, and 80GB is recommended.
  • Only tested on Linux.
  • Low-VRAM inference is very slow.
  • Inference time grows linearly with action length.

Topics

#video-generation#diffusion-model#game-ai#interactive-video#pytorch#generative-ai

FAQ

What is Hunyuan-GameCraft-1.0?
Open-source interactive game video generation model that turns a reference image and keyboard or mouse inputs into controllable gameplay-style video. Hunyuan-GameCraft is a video generation model from Tencent Hunyuan aimed at interactive game footage. Given a reference image, a text prompt, and a sequence of keyboard or mouse actions such as W, A, S, D, the model produces a video that follows those camera movements. Keyboard and mouse inputs are mapped into a shared camera representation space so that movements can be interpolated smoothly.
Is Hunyuan-GameCraft-1.0 free to use? What license does it have?
Hunyuan-GameCraft-1.0 is open source under the Other license. Check the license terms before using it in commercial products.
Is Hunyuan-GameCraft-1.0 still maintained?
No commits for 6+ months. The last commit to Hunyuan-GameCraft-1.0 was on Nov 28, 2025.
What can I use Hunyuan-GameCraft-1.0 for?
Researching controllable, action-conditioned video generation for games. Prototyping game-style clips from a single concept image and movement inputs. Experimenting with long-horizon video extension on multi-GPU Linux servers.
How do I install Hunyuan-GameCraft-1.0?
Clone the repository: git clone https://github.com/Tencent-Hunyuan/Hunyuan-GameCraft-1.0.git Create the conda environment: conda create -n HYGameCraft python==3.10 Install PyTorch and pip dependencies: python -m pip install -r requirements.txt Download the pretrained model weights (see weights/README.md) Run the Gradio web interface: bash scripts/launch_app_sp.sh
Where can I get Hunyuan-GameCraft-1.0?
The source code and installation instructions are on GitHub: https://github.com/Tencent-Hunyuan/Hunyuan-GameCraft-1.0

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects