BYOKey ElevenLabs Sound Effects

Once video generation got good, the obvious gap showed up: your clip has no sound. BYOKey_ElevenLabs_SFX is the tiny, focused node that fixes that - it turns a text description into a sound effect via ElevenLabs' sound-generation API and returns a real ComfyUI AUDIO output. Rain on a window, a whooshing camera move, a door slam, crowd ambience: type it, get it, drop it under your video in the same graph. The KB's audio essay frames exactly this as the "sound that matches a video (foley)" job, and it's the newest and thinnest layer in the ecosystem - which is why there aren't many nodes doing it, and this is one of them.

It's the sibling of the pack's ElevenLabs TTS node, and it's deliberately minimal: one text prompt in, one sound out, no voice to configure. If you're scoring a generated clip this is the fastest way to get from "silent video" to "has ambience."

How it works

It POSTs to ElevenLabs' /sound-generation endpoint with your description and settings, then decodes the returned audio bytes into ComfyUI's AUDIO output (waveform + sample rate). One call, one sound. The model list is currently a single entry - eleven_sfx_v2 - so the interesting knobs are the parameters, not the model choice.

The inputs that matter

• text - the description of the sound effect. This is the whole job. Be specific: "heavy wooden door slowly closing with a creak" beats "door."
• duration - 0.5 to 30 seconds, default 5. Realistic SFX are short; longer isn't better, it's just more of whatever the model decided the sound is.
• prompt_influence - 0 to 1, default 0.3. This is the dial that decides how literally the model follows your text. High values hug the description (and can get weird if your prompt is overly specific); low values produce more generic, safe sound. The 0.3 default is a sensible middle - nudge it up when the model keeps ignoring what you asked for.
• loop - when true, asks for a smoothly looping sound effect. This is the underrated one for ambience: a looping rain or room tone that never clicks is exactly what a background layer wants.
• output_format - mp3_44100_192 or opus_48000_192.

Output is audio (AUDIO) - wire it into a Save Audio node or combine it with your video downstream.

Installing it

It's one of 37 nodes in the comfyui-byokey pack, with no extra dependencies beyond ComfyUI's bundled torch/aiohttp/PIL/numpy/av. Install via ComfyUI Manager (search "byokey") or:

cd ComfyUI/custom_nodes
git clone https://github.com/MeteorAndy/comfyui-byokey.git

Restart, and it's under api/byokey/audio.

Gotchas

The honest limitations: ElevenLabs' sound effects are good at short, defined sounds and noticeably weaker at sustained, complex scenes - a 20-second "city street at rush hour" will sound like a loop of a small idea, not a real street. Keep effects short and layer a couple if you need a richer bed. Also, loop and long duration together can produce audible seams despite the loop flag - worth a listen before you ship it. The standard pack hygiene applies: your key is saved into the workflow JSON, so scrub it before sharing workflows. If your videos need a sound layer and you're already paying for ElevenLabs, this is the shortest path from silent clip to ambience.