ElevenLabs Pro - Sound Effects

Every video needs sound effects, and nobody wants to hunt through a royalty-free library for "door creak, horror, close mic." ElevenLabsPro_SFX is the pack's text-to-SFX node: describe the sound, get an audio clip back. The interesting part is that it ships with the correct model ID wired in, and that's more of a story than it sounds like.

Here's the history, because it explains why this node is worth trusting. The first version of the pack sent eleven_sfx_v2 as the model ID - which is not a model the live API accepts. Every SFX call silently failed or degraded until the v2.1 audit corrected it to eleven_text_to_sound_v2, verified against ElevenLabs' docs. The dropdown still shows eleven_sfx_v2 as a second option, but it's explicitly labeled as a legacy alias kept for backward compatibility. Default it stays on the correct one.

Mechanically it POSTs your text description to ElevenLabs' /v1/sound-generation endpoint with the model, duration, and influence settings, then decodes the audio back into an AUDIO dict. The inputs that actually matter:

• text - the sound description. Specificity pays off: "metallic door slams shut, with echo" beats "door."
• duration - 0.5 to 30 seconds. The node validates the range against the API's real limits, so no more out-of-bounds 422s.
• prompt_influence - 0 to 1, default 0.3. How tightly the result follows your prompt. Lower gives the model more creative room, which for SFX often sounds better than a rigid literal reading.
• loop - makes a seamlessly loopable effect, v2 model only. The sleeper feature for game developers and background ambience.
• seed, output_format - the usual suspects, with the usual Creator-tier caveat on mp3_44100_192 and opus.

The single output is audio. It's an InputCacheMixin node like the rest of the paid set, so identical inputs are cached - re-queueing won't re-bill, but any tweak to the prompt is a fresh charge.

Install is the pack standard:

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

Restart ComfyUI, or install "ComfyUI-ElevenLabs-Pro" via ComfyUI Manager. Just requests and soundfile; the generation happens in the cloud, so there's nothing to download.

Where people get burned: SFX is billed per generation, and short, low-influence clips multiply fast when you're building a library - run the batch, pick the keepers, and use seed to reproduce the ones that worked. loop=true is the trap for beginners who assume it loops the audio in the graph; it doesn't - it asks the model to generate a naturally loopable sound, which is subtly different and worth knowing before you queue a batch expecting an instant seamless loop. And as with all paid nodes here, check your tier on the format picker before you queue, or the call fails on permissions rather than content.