JoyAI‑Echo introduces a cross‑modal audio‑visual memory bank, a three‑stage post‑training pipeline, and a Director Agent to enable consistent, high‑quality, real‑time generation of minute‑level videos, achieving up to 7.5× inference speedup and state‑of‑the‑art benchmark scores.
OmniRoam introduces a panoramic, coarse‑to‑fine framework that generates long, trajectory‑controlled videos with higher spatial consistency and temporal coherence, offering a stable and controllable alternative to short‑clip generation and supporting real‑time preview, high‑resolution refinement, and 3D reconstruction applications.
The article details how JD’s JoyStreamer and JoyStreamer‑Flash models overcome text‑control weakness, multimodal conflict, and identity drift to enable long‑duration, free‑state, real‑time interactive digital‑human video generation, surpassing current SOTA models in benchmark scores and reaching 30 FPS inference speed for e‑commerce live streaming.
