ElevenLabs - Music Generation

Video needs a soundtrack, and most ComfyUI users don't have a band. This node takes a text description and returns a music track - actual music, not a loop - generated by ElevenLabs and delivered as an AUDIO output ready for your timeline. It calls the /v1/music endpoint, the same model behind the company's music features.

Only two required inputs: api_key and prompt. Describe the music in natural language - "lofi hip hop beat, warm, mellow, 80 bpm," "tense orchestral build with strings and timpani," "upbeat synthwave with driving bass." The model does more with a mood-plus-instruments-plus-tempo description than with a single genre word.

The dials:

• duration_seconds - 5 to 300, default 30. Longer tracks cost more, so start short to audition the vibe, then go long for the final.
• force_instrumental - true for no vocals. The single most useful toggle here; vocals in generated tracks are the fastest way to ruin a background score, so flip this on for anything under dialogue.
• seed - reproducibility (0 = random).
• output_format - the standard menu, with the Creator-tier note on high-bitrate mp3/opus.

One output: audio.

Where it fits

Score-for-video is the headline use: generate a bed, loop it or let it run under a scene, duck it under the voiceover. It's also genuinely good for mood boards and trailers, where you need a piece of music that sounds intentional. Because it's metered and per-call, the workflow that works is: short instrumental draft → listen → adjust the prompt → regenerate the final at full length. The pack re-runs every API node on each queue (IS_CHANGED), so regenerating is just a matter of changing the seed and queuing again.

Installing it

Part of the ComfyUI API Toolkit pack. Manager: search "API Toolkit". Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/IxMxAMAR/ComfyUI-API-Toolkit
cd ComfyUI-API-Toolkit
pip install -r requirements.txt

Restart. Needs requests and soundfile.

Gotchas

• This is the most expensive node in the pack per run - full tracks burn credits fast, and 300-second generations are real money. Audition at 30 seconds, commit at full length.
• The output is a single generation, not stems. You can't ask for the vocals separately; force_instrumental is your only mix control.
• If you need unlimited or local music, Stable Audio and similar open options are the counterweight - but for "give me a usable track from a sentence," this is the one.