JD Open-Sources JoyAI-EchoWM, a World Model with Coherent Audio-Video Navigation
jiqizhixin · x · 2026-09-19
JD Explore Academy unveiled and open-sourced JoyAI-EchoWM at JDD 2026, an interactive audio-visual world generation model.
Current world models produce stunning video but break when users actually step inside: change your route or camera angle and footsteps, ambient sound, and voices stop staying coherent with the visuals. First-person and third-person controls are fundamentally different, and any gap shatters immersion.
EchoWM unifies control across first-person (camera moves with the observer) and third-person (camera follows subjects with motion coordinated to composition) perspectives, and generates synchronized audio—footsteps, environment sounds, speech—that remains coherent as the user navigates. Built on JoyAI-Echo's native audio-video generation, it creates a truly enterable, controllable, explorable world.
More from Multimodal
- ComfyUI-NodeSnapshots: 2-3x frontend FPS boost via caching — External_Quarter · 2026-09-19
- Claymation Prompt Share: One Template for Adorable Tiny Clay Characters — azed_ai · 2026-09-19
- Watch: A Three-Clip Robot Short Made with MiniMax H3, Full Prompts Shared — JustLookingForNothin · 2026-09-19
- Image-to-video in 15 minutes: a 3-step Magnific workflow behind a 60M-view video — techhalla · 2026-09-19
- Ghost Visit: An AI Short Made with MiniMax H3 — apoke890 · 2026-09-19
- Generating a 40s motion graphic video with GLM writing code, no video model — 9r4n4y · 2026-09-19