EmptyLTXVLatentVideo_motorway_edition

LTX-Video (Lightning Tencent eXchange) made its name as the fast, cheap video model - real-time-ish video on a single consumer card, which back in early 2025 was a headline. What's less advertised is how weird its latent space is. Where most video models use a modest 12–16 latent channels, LTX-Video uses 128. Its spatial downscale is 32×, not 8×, and its temporal compression is 8×. This node builds an empty latent in exactly that shape, so the model doesn't get handed a tensor it can't make sense of.

The _motorway_edition wrapper from agilly1989/ComfyUI_agilly1989_motorway stores that blank latent on the Motorway pipe under a key, instead of returning it as a socket.

The inputs

• width (default 768, min 64, step 32) and height (default 512, min 64, step 32) - frame size. The 32-pixel step is LTX's strong spatial compression showing up as a real constraint: 768×512 is the friendly default, 1280×768 is a common stretch for 16:9.
• length (default 97, step 8) - frames. The temporal axis stores (length - 1) // 8 + 1 frames. 97 frames is roughly a 4-second clip at 24fps, which matches LTX's sweet spot for coherent motion.
• batch_size (default 1) - clips per batch.
• OUTPUT_LATENT_key (default LATENT) - the key the blank latent lands under on the Motorway.

Output is the MOTORWAY pipe; a motorway-ed sampler picks the latent back up via its INPUT_latent_key.

Why the shape matters

If you've only used image latents, the 128-channel thing sounds like a typo. It isn't - LTX's VAE trades channel count for a very tight spatial compression, and the sampler needs that exact layout. It's also why you shouldn't swap this node for a generic EmptyLatentImage "to save memory": the model will either error or produce mush, because the tensor is the wrong shape. Same reason each video model in this pack gets its own Empty* node rather than sharing one.

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 real download is the LTX-Video checkpoint plus its dedicated VAE from models/checkpoints and models/vae.

Pack-level caveat, same as every node in here: v1.1.7's own pyproject.toml reads "HOTFIX - REMOVED MONKEYPATCH AND GENERATED NODES" and __init__.py has the clone registration commented out, so this _motorway_edition node may not appear in your menu even though the core Motorway nodes work. The author's openly beta about it - "if things break it's because I broke it." If the node's missing, core EmptyLTXVLatentVideo does the same job and you can park its output on the bus with a Motorway 1x0 ramp.