Cloud Fetch Audio

Most nodes in this pack are promises. Cloud Fetch Audio is where the promise gets paid. It's a terminal node: it takes a CLOUD_AUDIO handle, appends a SaveAudio node to the accumulated workflow, submits the whole thing to Comfy Cloud, polls until the job finishes, downloads the audio file, decodes it, and hands you a real AUDIO object you can use in local nodes.

The audio path here is mostly the LTX 2.0 story. LTX 2.0 generates audio and video together, and the pack's LTX audio nodes produce a CLOUD_AUDIO handle from the cloud-side audio VAE decode. That handle eventually lands here, gets downloaded (the pack saves it as FLAC cloud-side), and is decoded into ComfyUI's AUDIO dict - a waveform tensor plus a sample rate - so you can preview it, feed it into an audio editor node, or pass it to Cloud Create Video as the soundtrack for your generated frames.

The inputs that matter

• audio (CLOUD_AUDIO) - the handle from an LTX audio decode or similar cloud audio chain.
• filename_prefix - default cloud_fetch_audio; sets the saved filename so you can keep runs apart.
• poll_interval (seconds, default 3) - how often it checks job status. Leave it.
• timeout (seconds, default 1800) - the kill switch. Bump it for long audio jobs; if the job isn't done in time you get a clear timeout error, not a hung graph.

Outputs: audio (AUDIO) - the decoded dict, ready for local nodes; prompt_id (STRING) - the cloud job ID, handy for matching console logs to runs and debugging in the platform's dashboard.

How it works

The terminal nodes all share one engine: submit to /api/prompt, poll /api/job/{id}/status and print each state change (queued_waiting → allocated → preparing → executing → success), then pull the file with /api/view and reconstruct a native ComfyUI type. For audio that means decoding the downloaded FLAC into the AUDIO dict via torchaudio - which is worth knowing, because torchaudio is not in the pack's own requirements.txt. In practice any ComfyUI install with torch has torchaudio, so it usually "just works," but it's the kind of hidden dependency that bites on a stripped-down setup.

How to install

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. ComfyUI Manager works too. If audio decoding fails with an import error, check that torchaudio is present in your ComfyUI environment.

Common issues

• Timeout on long generations - the default 1800s (30 min) is generous but not infinite. For big audio+video jobs, raise it preemptively; the failure mode is a timeout error, and you've then paid for a job whose results you can't get from the node (you can still fetch them from the platform by prompt_id).
• Import error on decode - the torchaudio dependency isn't declared in the pack's requirements. pip install torchaudio (matching your torch version) fixes it.
• PoC caveats - audio support is the newest, least-tested corner of a pack the author calls "very hacked together." The LTX 2.0 example workflow is your friend; stray from it and expect rough edges.