Cloud LTXV Concat AV Latent

LTX-2's audio-video trick only works if the video and audio live in one place. Before you sample, you have two separate latents - the frames from Cloud Empty Latent Video, the soundtrack from Cloud LTXV Empty Latent Audio. Cloud LTXV Concat AV Latent is the node that merges them into the single joint AV latent the model actually samples, contributing an LTXVConcatAVLatent step to the cloud workflow JSON.

Think of it as the cloud version of the concat node you'd use in a local LTX-2 template. It's one of those unglamorous plumbing nodes that does exactly one thing and does it quietly: two latents in, one latent out, and suddenly your cloud sampler is doing video with audio instead of silent film.

Inputs

• video_latent - a CLOUD_LATENT carrying your frames.
• audio_latent - a CLOUD_LATENT carrying the audio.

That's the whole list. Output is one latent (CLOUD_LATENT), which you feed into your Cloud KSampler / custom sampler as samples.

A practical note on sizing: the audio latent's duration comes from the frame count and frame rate you gave Cloud LTXV Empty Latent Audio, so the two halves have to agree on timing or the joint latent is incoherent. If your clip's audio drifts from the video, check those numbers first - it's the #1 mismatch in this chain.

Installation

Standard pack install:

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

Copy config.json.example to config.json, add an API key from https://platform.comfy.org/profile/api-keys, restart. In ComfyUI Manager by pack name; deps are requests, Pillow, safetensors.

Gotchas

• After sampling you have to split again. The sampler returns a joint AV latent, and you need Cloud LTXV Separate AV Latent to pull the video and audio halves back apart for decoding. Concat and Separate are bookends; forgetting the return trip is the classic mistake.
• This node is cloud-side only - it just builds JSON. The actual sampling still happens in your terminal sampler node, which is what polls Comfy Cloud.
• Proof-of-concept pack, and the AV chain is its newest territory. The repo's example_LTX_workflow.json shows the intended order (empty latents → concat → sample → separate → decode); mirror it with Cloud nodes and you're on solid ground.