AD_LTX_audio_input

LTX-2 and LTX-2.3 generate video and audio together in the same model - that's the headline feature that made lipsync and music-video work the killer app for the LTX line. But sometimes you don't want the model generating audio from scratch; you already have a voice track, a song, or a sound bed and you want the video conditioned on that. AD_LTX_audio_input is the node that encodes an existing audio clip into the audio-latent format LTX's sampler and config nodes expect, so your own audio drives the generation instead of the model inventing one.

How it works

It runs your audio through audio_vae to encode it into latent space, aligned to the frame timing you specify (frame_rate, frames_number, start_index). The result is a drop-in replacement for the empty audio_latent/audio_vae pair that AD_LTX_load_model outputs by default - wire this node's output in instead, and the rest of the LTX_video chain treats it exactly the same.

The inputs and outputs that matter

• audio (AUDIO) - your source clip.
• audio_vae (VAE) - the audio VAE to encode with, matching whatever's paired with your LTX checkpoint.
• start_index (default 0) - where in the clip to start encoding from.
• frame_rate (default 24) and frames_number (default 97) - should match the video-side settings from your AD_LTX_load_model/AD_LTX_config, since audio and video are meant to line up frame-for-frame in this pipeline.
• audio_denoise_scale (default 0, range 0–1) - adds noise before encoding. At 0 you get a clean, faithful encode; push it up if you want the sampler to have more room to deviate from your source audio rather than reproducing it closely - the same idea as denoise strength in an img2img workflow, applied to sound instead of pixels.

Outputs: audio_latent (LATENT) and audio_vae (VAE) - feed both straight into AD_LTX_sampler.

How to install it

Via ComfyUI Manager: search "ComfyUI-Apt_Preset". Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/cardenluo/ComfyUI-Apt_Preset.git

Then install.bat (Windows) or pip install -r requirements.txt (Linux/Mac), and restart. This node itself needs no extra downloads - it uses whatever audio_vae you feed it, which should be the one that ships with your LTX checkpoint.

Common issues & troubleshooting

Video and audio drift out of sync. Double-check frame_rate and frames_number here match exactly what you set on AD_LTX_load_model/AD_LTX_config for the video side. A mismatch is the most likely cause of the two tracks landing at different lengths.

Result barely resembles your source audio. audio_denoise_scale is probably too high - if you want the output to closely follow what you fed in, keep it near 0. If it's already near 0 and you're still seeing drift, that's normal generative variance from the sampler rather than something this node controls.

Not sure which audio_vae to use. Use the one that came bundled with the same LTX checkpoint you loaded in AD_LTX_load_model - mismatched VAEs across a pipeline is a common source of garbled output in any ComfyUI workflow, not just LTX.

Clip is longer than your target video and you don't know which part gets used. That's what start_index is for - set it to pick where in the source clip the encode window begins, rather than always encoding from the very start.