AceStep Generator (S42)

I'm going to be straight with you before you build a workflow around this one: right now, AceStep Generator outputs silence. Not quiet music, not a placeholder tone - a tensor of literal zeros. The class is a stub, the docstring calls it a "native integration bridge for AceStep," and the code behind it fills a stereo waveform with torch.zeros and returns a latent made of zeros too. If you were hoping this was the "local Suno" node, it isn't there yet.

ACE-Step, for the record, is the real thing: the open-weights music generator from ACE Studio and StepFun that this community adopted as the local answer to Suno, four minutes of music from a prompt in well under a minute on decent hardware (see the ace-step panel in our KB). The pack author clearly wants to bridge it into ComfyUI's graph - that's what this node is for. It's just not finished.

Here's what it actually does today, and it's worth understanding because it tells you where the author is heading:

• prompt - a multiline string, default "A gritty, rhythmic backing track." Sent nowhere yet, but it's the input a real AceStep call would take.
• duration_seconds - 5 to 120 seconds, step of 5. The stub does use this: it sizes the output to sample_rate * duration_seconds.
• seed - accepted, currently does nothing meaningful.

The outputs are the interesting part. It returns AUDIO and LATENT. The AUDIO is a 44.1 kHz stereo waveform (again: all zeros). The LATENT is shaped (1, 128, duration * 86) - and that 128 channels × 86 steps per second shape is the same audio-latent format the pack's other latent-audio nodes expect. So the design intent is: generate → latent → mess with it → decode, all in the graph. The plumbing exists, the engine doesn't.

That means the honest use today is as a stand-in. Wire it where a real generator will eventually sit, keep the graph structurally correct, and swap it out when the bridge lands. If you connect that LATENT to S42_AudioLatentDecoder, you get silence back - which is at least a round-trip that proves your latent plumbing works.

Installing it

This is a S42-CutFlow node, so installing it is installing the pack. ComfyUI Manager: search "S42 CutFlow", click Install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/GeekyGhost/S42-CutFlow.git
pip install -r S42-CutFlow/requirements.txt

On Windows portable it's .\python_embeded\python.exe -m pip install -r ComfyUI\custom_nodes\S42-CutFlow\requirements.txt. The requirements file is a single line - opencv-python-headless - and the audio nodes lean on torchaudio, which ships with ComfyUI. After restart you'll see the "[S42 CutFlow] Loaded ... total nodes" line in the console.

Where this leaves you

Treat it as a placeholder with clearly-marked inputs, not a generator. The pack's README is upfront that the whole suite is a work in progress being refined for eventual porting to LTX Desktop, and this node is the most unfinished thing in it. If you need music now, go run ACE-Step directly; this node will (hopefully) be the on-ramp once the author finishes the bridge.