JD.com introduces advanced AI models for interactive video generation

At the JD.com Global Tech Explorer Conference on September 9, the JD Explore Research Institute unveiled two new developments in the JoyAI-Echo series: the upgraded JoyAI-Echo1.5 long video generation model and the newly released interactive audio-visual world model EchoWM. JoyAI-Echo1.5 can generate audio-visual content lasting over 10 minutes, using the audio-visual Memory Bank architecture to maintain character images, voices, and story lines during multi-camera, long-duration generation. The model’s inference speed was increased by up to 7.5 times, allowing for real-time 1:1 generation at 480P resolution. EchoWM, on the other hand, integrates native audio-visual generation with user control, enabling AI content to transition from 'watchable' to 'enterable and exploratory.' EchoWM combines the strengths of two traditional approaches, generating 720P video, ambient sounds, music, and voice, with sound synchronized to scene, camera, and subject movements. It supports first-person navigation, third-person camera and subject collaboration control, and multi-round continuous exploration. In the WBench Navigation evaluation, EchoWM and EchoWM-Flash ranked first and second, with EchoWM securing the top position across 158 multi-round navigation cases. JD.com also demonstrated the new audio-video foundation model JoyAI-Video, targeting e-commerce advertising, short drama creation, and embodied intelligence training, with an official release planned for October.