Recently, Tencent unveiled a new generative model called Hunyuan-GameCraft. It is used to create gameplay videos where the movement of objects is controlled by the viewers.
Based on the footage demonstrated by Tencent, Hunyuan-GameCraft can generate a game scene from a text prompt with views from either the first or third person perspective. For example, a sea with a sailboat riding the waves or a track with a racing car. Users can control the movement of the boat/car using a keyboard and mouse.
In addition to generating gameplay videos, Hunyuan-GameCraft can be used to create "realistic" videos across a variety of genres, such as nature scenes with the ability to control the camera.
The model is trained on a million gameplay recordings from over 100 AAA games, including Cyberpunk 2077, Red Dead Redemption, and Assassin’s Creed.
According to the company, Hunyuan-GameCraft "significantly outperforms existing models" in a similar category. These likely include Veo 3 from Google and WHAMM from Microsoft.
Several examples: