ElevenLabs - Sound Effects

You've got a video and it's silent. You don't have a foley library, and you don't want to buy one. This node takes a text description and returns a matching sound effect as an AUDIO output you can lay directly into your timeline. It calls ElevenLabs' /v1/sound-generation endpoint, which means the model runs in their cloud and you just provide the words.

Required inputs: api_key and text - the description. "Heavy rain on a tin roof," "a cork popping," "distant thunder," "mechanical whirring." The more specific you are, the better the result, same as any prompt-based generation.

The controls:

• duration - 0.5 to 30 seconds, default 5. Short effects like a whoosh or a thud want a few seconds; ambiences want the max.
• prompt_influence - 0 to 1, default 0.3. How closely the output follows the prompt versus what the model improvises. Crank it when the first pass is too far from what you described.
• loop - ask for a smoothly looping effect, which is exactly what you want for ambience and background beds.
• model - currently just eleven_sfx_v2.
• seed - reproducibility (0 = random), plus the usual output_format menu.

One output: audio.

Where it fits

The obvious use is video: foley for a clip you generated with a video model, a sting for a transition, ambient room tone behind a voiceover. The loop toggle makes it genuinely useful for beds - set 30 seconds of "cafe murmur" and loop it under a scene. It's also a decent iteration tool: describe an effect, listen, tweak prompt_influence, regenerate. Because each queue re-runs (the pack sets IS_CHANGED on all API nodes), changing the seed and re-queueing gives you fresh takes every time.

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

• Like every hosted generator, it's metered and your prompt leaves the machine. For a handful of effects that's the right price; for a whole library, free local options exist but rarely match the quality per prompt.
• Short durations can come back thin - a 0.5 second whoosh is a coin flip. Give the model room to work.
• Sound is ComfyUI's shallowest layer; the AUDIO type is less universally supported than IMAGE, so if your save/preview node doesn't take AUDIO, check that you've got one that does.