Sound Notify (Inteliweb)

Sound Notify (Inteliweb) is a terminal node that plays a chime in your browser the moment execution reaches it. That's the whole job, and it's honestly the right job: when you queue a batch of 40 images and wander off, you want the graph to ping you when it's done instead of making you check the progress bar every ninety seconds. It's part of the Inteliweb node pack, which is mostly workflow plumbing - a seed node, Set/Get reroutes, a LoRA stack, a resource monitor - and this is the pack's one little bit of audio.

The reason it's a "workflow" node and not a "done" button is ComfyUI's execution model. ComfyUI only re-runs nodes whose inputs might have changed, and it works backward from its output nodes. A node that just wants to make noise has no output worth caching, so the author gave it the standard trick: InteliwebSoundNotify is an output node whose IS_CHANGED returns NaN, which forces it to fire on every queued run - even when everything feeding it came back from cache. Its changelog says exactly that: it re-runs on every launch even when upstream nodes use cached results. For a terminal node there's nothing downstream to poison, so the always-rerun cost is zero; this is one of the few places that trick is pure upside.

When the node runs, it emits a small ui payload saying which sound to play. A frontend extension listens for the executed event, then plays an MP3 from the pack's own assets/sounds folder. The five files - Tada, Tidi, Vista, Echo, and Dang - are served through an allow-listed route (/inteliweb/api/sound) that only ever hands out those bundled files, so it works on local installs, remote boxes, and behind proxies without exposing arbitrary paths. The sound comes out of the browser tab that has ComfyUI open, not the server.

The inputs, all three of them

• anything - a wildcard * input. Connect your final IMAGE, LATENT, MODEL, whatever. The value is ignored; it only exists to schedule the notification after that branch of the graph finishes. That's what makes it "this workflow is done" instead of "something ran."
• enabled - a simple on/off toggle (default on), so you can leave the node in the graph but silence it. More useful than it sounds when you're iterating on a prompt six times a minute and only want the chime on full batches.
• sound - the dropdown of the five bundled MP3s, Tada.mp3 by default.

That's it. No outputs, nothing downstream - it's a dead end by design.

Installing it

It ships with the whole Inteliweb pack, so there's no separate install:

cd ComfyUI/custom_nodes
git clone https://github.com/maoper11/comfyui_inteliweb_nodes.git
cd comfyui_inteliweb_nodes
python -m pip install -r requirements.txt

Or just search ComfyUI_Inteliweb_nodes in ComfyUI Manager and restart. The requirements.txt is two lines - psutil and nvidia-ml-py - and both are for the pack's Resource Monitor / System Check, not this node. No model downloads, no heavy deps, and it plays nice with both the Classic frontend and Nodes 2.0.

Where people get burned

The sound happens in the browser, so if your ComfyUI is running headless on a server and you're not staring at that tab, you will hear nothing. Remote users, welcome to the reason the changelog stresses that "the sound is served to the browser that has ComfyUI open."

Browser autoplay can also quietly eat it. If the page hasn't registered a click or keypress, some browsers refuse audio.play() - the code catches the rejection and just logs a warning to the console. In practice you've clicked Queue by the time anything runs, so it's rarely an issue, but if you hear silence and expect sound, check the browser console before blaming the node.

And remember the always-rerun behavior cuts both ways. It fires on every run, cached or not, so in a rapid-iteration loop it ticks like a metronome. That's what enabled exists for. If you want a "whole batch finished" chime rather than a "one image finished" tick, hang it off the last node in the chain, not the sampler.