JD.com has open-sourced JoyAI-Echo, a long audio video generation framework.

JD.com launched the JoyAI-Echo long-video generation framework, which solves three major industry challenges—character consistency, voice stability, and generation speed—through four technological innovations: a cross-modal audio and video memory library, memory-driven post-training inference speed improvement of 7.5 times, Director Agent conversational editing, and lightweight real-time super-resolution. Real-world testing shows that the framework achieves high consistency between the character and voice in a 5-minute video, with a speech accuracy of 0.8646, leading across all metrics.