LTX2 Empty Audio Latent

LTX-2 is a multimodal model: it generates video and audio as one latent. That's great until you're building a workflow that only wants video, and some node downstream insists on an audio tensor being present anyway. LTX2EmptyAudioLatent exists to solve exactly that - it builds a placeholder audio latent (a zeros tensor) sized to match your video's frame count, so audio-requiring nodes stop complaining. That's the whole job, and for a node this small, "the whole job" is the right size.

How it works

It looks at your video_latent, reads the frame count out of its shape, and constructs a zeros tensor in the LTX audio-latent format: (batch, channels, audio_frames, width). The shape math is deliberately rough - it approximates audio frames as video frames (the 16kHz audio / 25fps video relationship roughly cancels out in latent space), and defaults to 64 audio-VAE channels and a mono width of 1. The output is a LATENT you can drop into the audio_latent slot of a combine or decode node.

The inputs

• video_latent (required) - any video latent; the placeholder is derived from it.
• audio_channels - default 64, the audio VAE's latent channels. Leave it unless you know your audio VAE differs.
• duration_multiplier - default 1.0; scales the audio frame count relative to the video. Useful if your downstream expects audio longer than the video, e.g. 2.0 for a music-video workflow that generates long audio with a short visual lead-in.

Output is a single audio_latent (LATENT).

Where it fits in the workflow

The pack's own README diagrams the audio-video flow, and this node is the "no audio pipeline" shortcut:

[LTXVConcatAVLatent] -> [LTX2SeparateAVLatent] -> [LTX2EfficientSampler] -> [LTX2CombineAVLatent] -> decode
                                                  ↓
                                          [LTX2EmptyAudioLatent] (when you only have video)

If LTX2SeparateAVLatent can't find an audio component in a latent, it creates one of these automatically via its create_empty_audio option - so the standalone node is really for when you're assembling the pieces yourself.

Installation

It's part of the kakachiex2/comfyui-ltx2-efficient pack, so one install gets it. ComfyUI Manager (search "LTX2 Efficient") or:

cd ComfyUI/custom_nodes/
git clone https://github.com/kakachiex2/comfyui-ltx2-efficient
cd comfyui-ltx2-efficient
pip install -r requirements.txt

Restart ComfyUI afterward. No models, no heavy dependencies - requirements.txt is only the optional pynvml.

The honest caveat

A zeros audio latent is genuine silence, not "the model's idea of silence." Some audio-aware pipelines handle an empty audio slot gracefully; others will produce whatever a model does when handed a blank audio track - occasionally weird. If your goal is actually generating sound, don't use a placeholder; run the real audio pipeline. Use this node when you're building video-only output and a node on the path wants an audio input to exist. It's a tiny utility, it does one thing, and it does it without pretending to be more.