Hunyuan-GameCraft - An interactive game video generation framework launched by Tencent Hunyuan

What is Hunyuan-GameCraft?

Hunyuan-GameCraft is a high-dynamic-range interactive game video generation framework jointly developed by Tencent's Hunyuan team and Huazhong University of Science and Technology. It unifies keyboard and mouse input into a shared camera representation space, enabling fine-grained motion control and supporting complex interactive inputs. The framework introduces a hybrid history-conditional training strategy, which can autoregressively expand video sequences, preserving game scene information and ensuring long-term temporal coherence. Based on model distillation technology, Hunyuan-GameCraft significantly improves inference speed, making it suitable for real-time deployment in complex interactive environments. The model, trained on large-scale AAA game datasets, demonstrates superior visual fidelity, realism, and motion controllability, significantly outperforming existing models.

Hunyuan-GameCraft's main functions

• High Dynamic Range Interactive Video GenerationIt generates highly dynamic, interactive game video content from a single image and corresponding prompts, supporting real-time user control based on keyboard and mouse input.
• Fine motion controlIt unifies standard keyboard and mouse input into a shared camera representation space, supporting complex and fine-grained interactive input, such as speed and angle control.
• Long-term video generationIt supports generating long, continuous video sequences, preserving historical scene information, and preventing scene crashes.
• Real-time interactionSignificantly improves inference speed, reduces latency, supports real-time interaction, and enhances user experience.
• High visual fidelityTrained on a large-scale AAA game dataset, the generated videos have high visual fidelity and realism, and are suitable for a variety of game scenarios and art styles.

The technical principles of Hunyuan-GameCraft

• Uniform Action RepresentationThe keyboard and mouse inputs (such as W, A, S, D, arrow keys, etc.) are mapped to a continuous camera representation space. The input camera trajectory is encoded into a feature vector based on a lightweight motion encoder, achieving smooth motion interpolation.
• Hybrid historical condition training strategyBased on historical context integration and mask indicators, video sequences are autoregressively extended. In each autoregressive step, historical denoised blocks are used as conditions to guide the denoising process of new noise latent variables, preserving scene information and solving the error accumulation problem in autoregressive generation.
• Model distillationBased on the Phased Consistency Model (PCM), model distillation is achieved, which freely guides the original diffusion process and classifier to distill into a compact eight-step consistency model, significantly improving inference speed and reducing computational overhead.
• Training on large datasetsThe model is trained on over one million game recordings from more than 100 AAA games to ensure broad coverage and diversity. Fine-tuning with carefully annotated synthetic datasets improves accuracy and control.

Hunyuan-GameCraft project address

• Project official websitehttps://hunyuan-gamecraft.github.io/
• GitHub repository: https://github.com/Tencent-Hunyuan/Hunyuan-GameCraft-1.0
• HuggingFace model libraryhttps://huggingface.co/tencent/Hunyuan-GameCraft-1.0
• arXiv technical paperhttps://arxiv.org/pdf/2506.17201

Official example from Hunyuan-GameCraft

In a single-action scene, the visuals are dynamic and natural, with the windmill rotating naturally.

• Prompt:A picturesque village scene featuring quaint houses, a windmill, lush greenery, and a serene mountain backdrop under a bright blue sky.
  ChineseA picturesque rural scene with quaint cottages, windmills, lush greenery, and a tranquil mountain backdrop under a bright blue sky.

Multiple action scenes allow for simultaneous switching of perspective and movement.

• Prompt:A sunlit courtyard features white adobe buildings with arched doorways and windows, surrounded by lush greenery and palm trees, creating a serene Mediterranean ambiance.
• ChineseThe sunlit courtyard, with its white adobe buildings featuring arched porches and windows, is surrounded by lush greenery and palm trees, creating a tranquil Mediterranean atmosphere.

Historical consistency; no distortion after the camera is moved away and then returned.

• Prompt:A medieval stone castle stands tall under a dark sky, its glowing windows contrasting with the surrounding snow-covered landscape.
• ChineseA medieval stone castle stands tall under a dark sky, its gleaming windows contrasting sharply with the surrounding snow-covered landscape.

Application Scenarios of Hunyuan-GameCraft

• Game video generationQuickly generate game trailers, demo videos, and in-game cutscenes to help game developers validate game concepts and designs in the early stages.
• Game testingIt automatically generates game scenes and interactions for testing game performance and player experience, reducing the workload of manual testing.
• Game content expansionGenerate new levels, scenes, and interactive content for existing games, extending the game's lifespan and increasing player engagement.
• Interactive video contentIt generates interactive video content for video platforms and social media, allowing users to input commands to control the video's progression and providing a completely new viewing experience.
• Virtual Reality (VR) and Augmented Reality (AR)Generate immersive, interactive content in VR and AR applications to enhance user immersion and engagement.