AnswersPublished17 sources
JD Open-Sources JoyAI-EchoWM and EchoWM-Flash for Interactive Audiovisual Worlds
JD announced open source JoyAI EchoWM and the four step EchoWM Flash at JDD 2026. A unified camera intent interface turns keyboard or controller input into continuous six degree of freedom trajectories for first person exploration and third person camera views.
Published byEdited with GPT-5.6 TerraImages generated with GPT Image 2
Research answer
Key takeaways
- JD announced open source JoyAI EchoWM and the four step EchoWM Flash at JDD 2026.
- A unified camera intent interface turns keyboard or controller input into continuous six degree of freedom trajectories for first person exploration and third person camera views.
- The release is part of JD’s physical AI push, alongside plans for a 100,000 GPU domestic compute cluster and more than 10 million hours of real world activity video data.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What did JD.com announce at the September 10, 2026 JDDiscovery (JDD) conference about open-sourcing JoyAI-EchoWM—its interactive audiovisual. Article summary: JD announced that it is open-sourcing JoyAI‑EchoWM, an “enterable” interactive audiovisual world model, and EchoWM‑Flash, a faster four-step causal-streaming version. The release is best understood as part of JD’s physic. Topic tags: general, academic, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, char
JD’s core announcement at JDDiscovery (JDD) 2026 was the open-sourcing of JoyAI-EchoWM, an interactive audiovisual world model, alongside EchoWM-Flash, a lighter four-step causal-streaming version. Rather than generating a fixed video clip, the models are designed to let a user navigate a generated scene while video and sound evolve together. 1
6
What JoyAI-EchoWM is designed to do
EchoWM generates 720p video together with synchronized environmental sound, music, and speech. The intended experience is an “enterable” generated world: as the viewer changes direction or moves through a scene, the model updates both the imagery and audio rather than attaching a separate soundtrack afterward. 1
2
The system uses a unified camera-intent representation. It converts keyboard or controller commands into continuous six-degree-of-freedom camera trajectories, enabling first-person roaming as well as third-person and follow-style camera views. Camera paths condition the visual output, while scene events are meant to yield matching sounds—for example, footsteps, impacts, dialogue, or music. 1
11
EchoWM-Flash: the faster streaming variant
EchoWM-Flash is the four-step causal-streaming variant. Its purpose is to reduce sampling steps while retaining responsive interactive generation. In the paper’s 158-case WBench Navigation results, EchoWM scored 81.7 on average and EchoWM-Flash scored 81.0. EchoWM’s reported interaction and consistency scores were 87.2 and 89.8, respectively; Flash recorded an interaction score of 87.9. 1
Those results should be read as developer- and paper-reported benchmark findings, not as independent evidence that either model is ready for every real-time production use case. The separate WBench repository lists JoyAI-Echo-1.5 (WM) at 81.6 and its four-step Flash version at 81.0, which explains the small 81.6-versus-81.7 difference in coverage. 15
How it maintains an interactive scene—and where it falls short
EchoWM supports multi-turn exploration by conditioning generation on a short trailing audiovisual context window. That allows the model to continue a navigated sequence, but it is not the same as maintaining a persistent, explicit 3D representation of the world.
That distinction matters: without persistent 3D memory, longer explorations can accumulate spatial or visual drift. JD’s reported results show strengths in interaction and consistency on the benchmark, but the architecture’s context-based continuity remains a practical limitation for long-running navigation. 1
11
Training approach and public availability
The accompanying research describes a world-data engine and staged training pipeline using gameplay logs, human-play footage, Unreal Engine-based pose-grounded simulation, and internet video. The goal is to connect controllable camera motion, visual dynamics, and synchronized audio generation in one model. 1
JD’s paper links to the public code repository, jd-opensource/JoyAI-Echo, making the release relevant to researchers and developers experimenting with interactive world-model systems. 1
Why the release fits JD’s physical-AI strategy
EchoWM was presented as one component of a larger JD physical-AI effort spanning data, computing, models, and logistics operations. JD Cloud said it plans to build a domestically developed 100,000-GPU cluster for large-model training, inference, and embodied-AI workloads, and to collect more than 10 million hours of video showing real-world human activity. 18
22
JD Logistics’ Meta Brain 3.0 is the operational counterpart to that strategy. JD says the system coordinates decisions across warehousing, transportation, and delivery, reducing the time required to calculate optimal routes for hundreds of millions of parcels from minutes to seconds. 18
22
In that context, the EchoWM release is best understood as an open-model and research-infrastructure move for interactive and physical-AI development—not a financing announcement. Its most notable technical proposition is straightforward: combine controllable camera navigation with jointly generated video and sound, while being clear that long-horizon world consistency is still an open challenge. 1
11
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "JD Open-Sources JoyAI-EchoWM and EchoWM-Flash for Interactive Audiovisual Worlds"?
JD announced open source JoyAI EchoWM and the four step EchoWM Flash at JDD 2026.
What are the key points to validate first?
JD announced open source JoyAI EchoWM and the four step EchoWM Flash at JDD 2026. A unified camera intent interface turns keyboard or controller input into continuous six degree of freedom trajectories for first person exploration and third person camera views.
What should I do next in practice?
The release is part of JD’s physical AI push, alongside plans for a 100,000 GPU domestic compute cluster and more than 10 million hours of real world activity video data.