⭐ Star Sound Mixer

AI video now ships with generated soundtracks, and mixing a few audio sources - the model's own audio, a music track, a voiceover - used to be a "go open a DAW" task. Star Sound Mixer brings it back into the graph: it takes up to 12 audio inputs, gives each one its own volume slider right on the node, and sums them into a single AUDIO output. It's a mixing desk that fits in a socket.

How it works

The clever bit is the dynamic inputs. The node starts with one slot (audio_1 + volume_1), and each time you connect a slot, the next one appears - up to 12 channels. Every connected audio gets a per-channel volume slider (0–100%, default 100%). On run it does three things in order:

1. Resample every input to the first connected input's sample rate, so channels from different sources (say a 44.1 kHz music file and a 48 kHz LTX soundtrack) actually line up sample-for-sample.
2. Pad inputs of different lengths and channel counts with zeros so they sum cleanly.
3. Sum and protect - if the mixed signal would exceed 0 dBFS, it normalizes instead of clipping. That's the part that saves you from a blown-out, distorted mix when you've stacked a few loud sources.

Inputs and outputs

Inputs: audio_1 (the reference - it sets the sample rate for everything else), its volume_1 slider, and the dynamically appearing audio_2…audio_12 / volume_2…volume_12. Output: a single audio socket carrying the mixed stream. Wire it into a video encoder with save_audio on (Star Video Compressor and Star Slideshow Maker both accept it), or into an audio save node.

Installing

Part of StarNodes:

cd ComfyUI/custom_nodes
git clone https://github.com/Starnodes2024/ComfyUI_StarNodes
cd ComfyUI_StarNodes
pip install -r requirements.txt

Restart, search star under ⭐StarNodes/Video. ComfyUI Manager users: search "Starnodes".

Gotchas

The reference-sample-rate rule is the one thing to remember: whichever audio is in audio_1 defines the output rate, so if you want a 44.1 kHz final mix, put the 44.1 kHz source in the first slot. Everything else gets resampled to it - usually fine, but if you notice one track sounds slightly off-pitch or warbley, that's a resample artifact on a heavily-converted source, not a bug. And the normalizer only kicks in when the sum peaks past 0 dBFS; you can still end up with a loud, dense mix that's technically under the ceiling but sounds mushy, so keep individual volumes sensible rather than maxing every channel and relying on the safety net.