LTX-2 Fast

Most video nodes bury you in step counts and CFG values before you've even decided what the clip is about. LTX-2 Fast takes the opposite approach: the two parameters that would tempt you to overthink - duration and fps - are dropdowns with sensible presets, and everything else is optional. Pick 8, pick 25, type a prompt, queue. It's the "stop fiddling and make something" node of the LTX family, served from Runware's cloud.

This is the distilled, fast variant of LTX-2 (lightricks:2@1), a deliberately reduced control surface over the base model. The one thing it keeps that base LTX-2 doesn't expose here is the party trick: synchronized audio generation.

How it works

Standard pack mechanics - the node sends a videoInference request over REST via the runware-sdk and wraps the returned MP4 as a native VIDEO output. What makes this node distinct:

• duration - a preset dropdown: 6, 8, or 10 seconds, default 6. No arbitrary values.
• fps - 25 or 50, default 25. Pick 50 when you want slow-motion-worthy smoothness.
• providerSettings.lightricks.generateAudio - the LTX signature move, one toggle: generate synchronized audio for the clip. This is the thing Wan users keep asking for.
• positivePrompt is required; width / height default to 1024; frameImages (IMAGE) for image-to-video; numberResults 1–4.
• No steps, no CFG, no negative prompt, no advanced_json - the model doesn't want them, and the pack doesn't fake them.

The output

One video socket, type VIDEO, into Preview Video or Save Video. On an older ComfyUI build it degrades to a file path string - update ComfyUI for the native type.

Installing it

The pack install is shared across every node here:

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

Restart, set your key via ComfyUI Settings → Runware API key, RUNWARE_API_KEY, or runware auth login. Nothing heavy installs - SDK, pillow, soundfile.

Where people get burned

Not much to burn on - that's the design. The main trap is forgetting this is the base LTX-2 in fast form, not 2.3: if your renders come out soft or the prompt adherence disappoints, that's the older model's known character, and the pack's Runware_lightricks_ltx_2_3 node is the upgrade path. And generateAudio isn't a watermark - when it's on, the audio is genuinely part of the render, so don't blame the node for a louder receipt.