Cloud LTXV Empty Latent Audio

Every diffusion run needs a starting latent - some blank, noise-shaped canvas the sampler refines into the thing you asked for. Video has one, images have one, and in LTX-2's audio-video world, audio has its own: Cloud LTXV Empty Latent Audio creates the empty audio latent, sized by how much sound you want, and contributes an LTXVEmptyLatentAudio step to the cloud workflow JSON.

It's the audio sibling of Cloud Empty Latent Video, and it's the first node in the LTX-2 audio chain: empty audio latent → Cloud LTXV Concat AV Latent (joined with the video latent) → sampler → Cloud LTXV Separate AV Latent → decode. You need it whenever your clip is supposed to have a soundtrack.

The inputs that matter

• frames_number - INT, default 97. The length of the clip in frames. More frames = longer audio.
• frame_rate - INT, default 25. Combined with frames_number, this sets the actual duration (97 frames ÷ 25 fps ≈ 4 seconds). This is the number that has to agree with your video latent's frame rate.
• batch_size - INT, default 1. How many audio latents to make.
• audio_vae - a CLOUD_AUDIO_VAE handle from Cloud LTXV Audio VAE Loader. The empty latent is shaped in the audio VAE's latent space, so the VAE has to be loaded first.

Output is a single Latent (CLOUD_LATENT).

Installation

Standard for the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/Dobidop/ComfyUI-CloudAPI-worker

Copy config.json.example to config.json, paste an API key from https://platform.comfy.org/profile/api-keys, restart. Manager-installable as "ComfyUI-CloudAPI-worker"; deps: requests, Pillow, safetensors.

Gotchas

• Frame rate consistency, again. If your video latent is 30 fps and this node is set to 25, your audio duration won't match the clip and things drift. Check both latents.
• It's a JSON-building node like the rest of the LTX cloud family - nothing happens until the whole chain reaches a terminal node that submits the job.
• Proof-of-concept pack; the audio path is the newest part. The repo's example_LTX_workflow.json shows the exact ordering that's known to work, so use it as your wiring reference.