EmptyMochiLatentVideo_motorway_edition

Mochi 1 is Genmo's open video model - the one that made long, coherent motion feel achievable before LTX and Wan ate the local-video conversation. If you're building a Mochi workflow, this node makes the empty latent tensor the model denoises into a clip: 12 channels, an 8× spatial downscale, and a 6× temporal compression. The _motorway_edition version from agilly1989/ComfyUI_agilly1989_motorway wraps that core behavior so the blank latent is written to a named key on the Motorway pipe instead of handed out as a socket.

The inputs

• width (default 848, step 16) and height (default 480, step 16) - frame size in pixels. 848×480 is Mochi's comfortable default; 1280×720 is the usual upgrade if VRAM allows.
• length (default 25, min 7, step 6) - number of frames. The 6-step and 7-frame minimum are Mochi's temporal compression asserting itself: (length - 1) // 6 + 1 frames are stored, and anything under 7 frames just doesn't make a coherent clip.
• batch_size (default 1) - clips per batch.
• OUTPUT_LATENT_key (default LATENT) - the key the blank latent is stored under on the Motorway.

The only output is the MOTORWAY pipe. A motorway-ed sampler later reads the latent back through its INPUT_latent_key.

How it fits

Text-to-video from scratch: MotorwayStart → this node parks LATENT → a motorway-ed sampler consumes it and writes back a denoised latent under a new key → a motorway-ed VAE decode turns it into pixels. Nothing about the latent math changes vs. core EmptyMochiLatentVideo - the wrapper only reroutes where the result goes, and adds the key-naming ceremony that is the Motorway's whole personality.

One thing worth knowing if you're new to video: 25 frames at ~12fps is about two seconds, which is a normal Mochi generation, not a mistake. Long clips are a pipeline problem (chunking, interpolation, VACE-style joining) that no empty-latent node solves.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/agilly1989/ComfyUI_agilly1989_motorway

Restart ComfyUI, or install via ComfyUI Manager under "ComfyUI_agilly1989_motorway." No extra Python deps; the heavy lift is Mochi's checkpoint + VAE in models/checkpoints and models/vae.

And the standard pack-level caveat: v1.1.7's pyproject.toml calls itself "HOTFIX - REMOVED MONKEYPATCH AND GENERATED NODES," and __init__.py has the clone nodes commented out. The core Motorway nodes work; this _motorway_edition clone may not be in your menu until the author re-enables the generator. If it's absent, core EmptyMochiLatentVideo is a drop-in replacement - park its output on the bus with a Motorway 1x0 ramp and you've rebuilt the same thing by hand.