LTXVImgToVideo_motorway_edition

Image-to-video is the workhorse of local video generation: one still, and the model animates the rest. LTXVImgToVideo_motorway_edition is the LTXV flavor of that step, re-plumbed for the Motorway pipe. It takes positive and negative conditioning, a VAE, and a starting image out of the pipe, encodes the image into a video latent, and writes the conditioned positive/negative plus the latent back under keys.

LTX's whole identity is speed - it's the "draft in seconds" tier of local video, commonly used to block out motion before refining in a heavier model like Wan. This node is where an LTXV img2vid run actually begins: everything before it is setup, everything after it is sampling.

A few settings matter more than the rest. length is your total frame count - the default 97 is roughly 4 seconds at 25 fps, and it's a 8-step increment because of how the spatiotemporal VAE packs frames. width/height are generation resolution (steps of 32, defaults 768×512). And image_noise_scale (default 0.15) is the one worth tuning: per the tooltip it's the "amount of noise to apply on conditioning image latent." Low means the first frame hugs your input tightly; higher gives the model more room to drift away from the source. It's effectively your "how faithful is the start" dial. batch_size lets you generate several clips at once.

The wrapper

• MOTORWAY 🚌💨 - the pipe. Required.
• INPUT_positive_key (default positive), INPUT_negative_key (default negative) - conditioning to tag and pass through.
• INPUT_vae_key (default vae) - the VAE used to encode the starting image.
• INPUT_image_key (default image) - the starting frame.
• width (default 768, step 32), height (default 512, step 32), length (default 97, step 8), batch_size (default 1), image_noise_scale (default 0.15).
• OUTPUT_positive_key, OUTPUT_negative_key (defaults positive/negative) and OUTPUT_latent_key (default latent) - the conditioned pair and the video latent.

The output port is the pipe; pull latent out with a Motorway AxB ramp and feed it, with positive/negative, into your sampler to actually produce frames.

Installing

Part of the Motorway pack:

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

Or ComfyUI Manager → search ComfyUI_agilly1989_motorway → Install. No node-specific downloads - but LTXV checkpoints, a VAE, and core's LTXV model-sampling nodes are all required for a real run.

Gotchas

• Everything it reads must already be in the pipe: Key: 'vae' doesn't exist in motorway means the VAE never got written upstream, and so on for image and positive/negative. Check keys first, wiring second.
• Length math: keep length on the 8-step and remember LTXV renders efficiently at moderate sizes - cranking resolution is where speed-tier video starts to hurt.
• image_noise_scale is the knob beginners either ignore (and wonder why the first frame looks off) or max out (and wonder why the video abandoned the input). Default 0.15 is a sensible middle.
• Beta pack: author's README says "IF THINGS BREAK ITS BECAUSE I BROKE IT," and generated clones have been toggled off in some releases. Missing from the menu after install is a version thing.

This is the node that makes an LTXV img2vid run start inside the pipe. Get your image, VAE, and conditioning in, and this hands your sampler everything it needs.