WorldFoundry
English中文

JoyAI-Echo 1.5 / Echo-LongVideo

JoyAI-Echo 1.5 / Echo-LongVideo from JD.com (JoyAI / Echo Team): reference-to-video and audio-video-generation; Runner parity pending; dedicated `joyai-echo-longvideo`.

Edit this pageGitHubMarkdown

JoyAI-Echo 1.5 is an inference-only long-form audio-video generator. Its R2V request can combine a prompt, an optional first frame, and as many as seven ordered image/audio memory slots so identity, voice, and story context persist across independently editable shots.

Official project teaser: JoyAI-Echo-1.5 samples spanning interactive worlds and real-world human scenes.
Official project teaser: JoyAI-Echo-1.5 samples spanning interactive worlds and real-world human scenes.

Code: jd-opensource/JoyAI-Echo.

Build your run

Pipeline
JoyAIEchoLongVideoPipelineworldfoundry.pipelines.joyai_echo.pipeline_joyai_echo
TaskInteractive Video interactive-video
Environmentjoyai-echo-longvideo
DeviceCUDA 12.8
Runnerworldfoundry.pipeline
Loadingfrom_pretrained
Invocationnative_or_standard_invocation
worldfoundry-eval run \
  joyai-echo-longvideo \
  --pipeline.task-profile interactive-video \
  --json

What it is for

Use joyai-echo-longvideo for catalog text-to-video inference. The recipe exists so you can launch a recorded T2V profile without assembling an upstream demo.

The release builds on LTX-2.3 and adds few-step DMD generation plus a paired cross-modal memory bank. WorldFoundry keeps the official Python 3.11, PyTorch 2.8, CUDA 12.8 runtime external and communicates through the checked-in CLI contract. It is published by JD.com (JoyAI / Echo Team).

Catalog tasks include reference-to-video, audio-video-generation, and long-video-generation. Catalog aliases include joyai-echo, echo-longvideo, and joyai-echo-1.5. Documented modalities: image in, video out.

Recorded evaluation entry points include ChronoMagic-Bench (Long-horizon temporal continuity across generated shots), VBench (Baseline video quality and motion evaluation).

Runtime contract

Optional prompt, input_path, interactions, frames, and fps; artifacts generated_video and joyai-echo-longvideo.mp4.

CLI task profiles: interactive-video. WorldFoundry binds JoyAIEchoLongVideoPipeline.

Family status: Runner parity pending.

Recorded interactive-video defaults: frames=241, fps=25, height=736, width=1280, seed=42.

Clone JoyAI-Echo at the recorded revision, prepare its dedicated environment, and pass source_root plus local Echo and gated Gemma checkpoints. Generation is safety-gated behind execute=True. FP4 also needs NVIDIA ModelOpt; production voice-memory extraction needs the upstream MSST setup.

Running in WorldFoundry

Launch worldfoundry-eval run joyai-echo-longvideo --pipeline.task-profile interactive-video to sample the clip. The command builder above switches --pipeline.task-profile among recorded tasks and still emits worldfoundry-eval run.

Recorded weight repositories include jdopensource/JoyAI-Echo and google/gemma-3-12b-it. Prepare the environment and check local assets with bash scripts/inference/prepare_model_infer.sh joyai-echo-longvideo. Add --download to fetch missing public checkpoints.

Status: Runner parity pending.