Dynamic Insight: Beating AI News Flash, Tencent's General Manager of Polymorphic Multimodal Reinforcement Learning, Tianyu Pang, unveiled HyCreator and opened applications for early access.
HyCreator is defined as an agent harness for long video generation (an intelligent framework that connects models, tools, and production processes), capable of generating videos of approximately 10 minutes with zero manual intervention. Users can switch to real-time interactive editing at any point if adjustments are needed.
The official website has released a batch of real works, with the longest being 10 minutes and 27 seconds, as well as videos at 9 minutes 45 seconds and 9 minutes 59 seconds. Each work also allows viewing of the Agent's generation Trace. In a public case study, the Pro mode involved 13 steps, generated 45 intermediate outputs, performed quality checks on segments, transitions between shots, and overall coherence, before synthesizing the final output.
HyCreator offers two video creation modes: Lite and Pro. As of now, the underlying video models used, the total generation time, and costs have not been disclosed.

