TaoLiveAIGC introduced the TaoMate-H3 runtime for streaming video generation

The TaoLiveAIGC team has released TaoMate-H3 — a specialized runtime designed for low-latency streaming generation of audio and video. The system is based on the MiniMax H3 architecture and uses a 3-step LoRA adaptation method to optimize the output process.

The tool supports content generation in 480p, 768p, and 1080p resolutions. The technical implementation allows for inference on a single node using a configuration of 4–8 graphics processors. According to the developers' internal tests, the system achieves a score of 80 points when generating content for 1 hour.

The architecture is oriented towards working with long video formats, which is critical for real-time streaming tasks. The use of LoRA adapters allows for flexible customization of the model for specific visual tasks without the need for complete retraining of the base MiniMax weights.

The release is aimed at reducing the hardware entry threshold for creating high-quality streaming content. The ability to scale to 8 GPU allows for deploying the solution within local server capabilities to ensure a stable frame rate when generating video streams.