Echo-WM T8 nodes v0.1.0

Echo-WM T8 ComfyUI nodes v0.1.0.

• Real Echo-WM Flash Preview / four-step Causal generation.
• Native ComfyUI VIDEO output with an isolated Python 3.11 / Torch 2.9 runtime.
• Built-in guarded 256x128 smoke preset and 10-second low-resolution preset.
• Verified end-to-end in ComfyUI at 256x128, 25 frames, 24 fps.
• Matching gated model bundle: https://huggingface.co/t8star/Echo-WM-Comfy
• Comfy Registry package: echo-wm-t8

Model weights are intentionally not included in this source release and remain governed by their upstream licenses.

Real 10-second validation

• Executed through the released ComfyUI node path at 128x128, 241 frames, 24 FPS, and 4-step Causal Flash.
• Result: 10.042 seconds, H.264/yuv420p; all 241 decoded frame hashes are distinct.
• Action schedule: forward, left+forward, jump+forward, forward.
• Attached sample: echo-wm-t8-real-10s-128x128-h264.mp4, SHA-256 0e22bf53e316f4c80d1ef8a25c81d81ba3270501ff21a999592313967dd67ad3.
• A 256x128 / 241-frame attempt exhausted the safe RAM reserve on the 64 GiB validation host and was interrupted; use 128x128 for this hardware smoke test.

JoyAI-Echo Studio v0.1.0-beta.3

JoyAI-Echo Studio v0.1.0-beta.3 / Echo-WM T8 v0.1.0

This public beta combines the standalone Echo-WM T8 ComfyUI nodes with the
weight-free JoyAI-Echo Studio Windows portable bundle.

Windows signing notice: this beta is published without an Authenticode
signature. Windows or SmartScreen may display an unknown-publisher warning;
verify the SHA-256 and release assets before running it.

Verified paths

• Real ComfyUI graph execution: Echo-WM Flash/Causal, 4 steps, 256×128,
  25 frames at 24 fps, native VIDEO output.
• Full low-resolution rollout: Echo-WM Flash/Causal, 4 steps, 128×128,
  241 frames at 24 fps (10.042 seconds); all 241 decoded frames are distinct.
• Action schedule: forward → turn left → jump → forward.
• Action HUD output is enforced and re-probed as H.264; MPEG-4 Part 2 is not
  accepted as a silent fallback.
• Model installer resolves every Echo-WM file from the pinned, gated
  t8star/Echo-WM-Comfy revision
  9eda68f03820e58676f31947b4bb4b6e1f40e70c.
• 74 automated tests pass across the integration, node package, packaging,
  and desktop launcher.

On the 64 GiB validation host, a 256×128 / 241-frame rollout exhausted the
safe RAM reserve and was interrupted. Use 128×128 for the ten-second hardware
smoke test on comparable machines; 256×128 ten-second output is not claimed as
validated by this beta.

Downloads

The Windows archive is split into parts smaller than GitHub's 2 GiB per-file
limit. Download the .parts.json, every matching .partNNN file, and
Join-JoyEchoArchive.ps1 into one directory. A large numbered part may itself
have a transport-level .parts.json; reassemble those nested parts first, then
reassemble and verify the main ZIP:

Get-ChildItem -Filter '*.part*.parts.json' | ForEach-Object {
  powershell -ExecutionPolicy Bypass -File .\Join-JoyEchoArchive.ps1 `
    -IndexPath $_.FullName
}
powershell -ExecutionPolicy Bypass -File .\Join-JoyEchoArchive.ps1 `
  -IndexPath .\JoyAI-Echo-Studio-v0.1.0-beta.3-win32-x64.zip.parts.json

Expected reconstructed ZIP: 12,384,705,709 bytes, SHA-256
a6cf918dbdfd680b680a91d25dd7e43d98839f32593b05cd0bef8b84456cfe2e.

The real ten-second validation clip is attached as
echo-wm-t8-real-10s-128x128-h264.mp4; SHA-256
0e22bf53e316f4c80d1ef8a25c81d81ba3270501ff21a999592313967dd67ad3.

The portable archive deliberately contains no model weights. Request access
and download the required gated bundle from
t8star/Echo-WM-Comfy.

Repositories:

• Nodes: https://github.com/T8mars/Comfyui-Echo-WM-T8
• Models: https://huggingface.co/t8star/Echo-WM-Comfy
• Upstream project: https://github.com/jd-opensource/JoyAI-Echo

Review the included licenses and third-party notices before use. This release
does not add any prohibition on redistribution; upstream license terms remain
the governing terms.