Tencent On December 3, it was announced thatHunyuan Large ModelGo live and open source Vincennes video capabilities, with 13 billion participants, to support the ChineseBilingual English input.
Officials claim that Tencent's hybrid video generation large model can generate "Ultra-realistic "high quality videoThe screen generated is not easily distorted; in the mirror or mirror scene, the mirror reflection action and the outside can be completely synchronized.The reflection of light and shadow is basically in accordance with the laws of physics.
According to the introduction, the Tencent hybrid video generation of large modelsUtilizes DiT architecture,Adaptation of a new generation of text encoders to improve semantic compliance, better able to cope with multiple subject depictions and achieve more detailed instructions and picture presentation.
In "Tencent Yuanbao App", go to "AI Application" and select "AI Video" to apply for a trial.
Tencent said that the open source contains a complete model of model weights, inference code, model algorithms, etc.Free to use and develop eco-plugins for corporate and individual developers.