AceStep 1.5 SFT Save Audio

Every ComfyUI music workflow needs a way to get the audio out of the node graph, and AceStepSFTSaveAudio is the pack's take on it. Wire an AUDIO in, pick a format, and it writes the file with an auto-incrementing name. Nothing clever, nothing fragile - which is exactly what you want from an output node. It also renders an interactive waveform on the node itself, so you can audition the render before you even open a player.

What it does

Inputs are minimal:

• audio - the AUDIO to save, normally the audio output of the Generate node.
• filename_prefix (default audio/AceStep) - prefix for the filename; you can include a subfolder path, so audio/AceStep saves into ComfyUI/output/audio/ and the node auto-increments: AceStep_00001_.flac, AceStep_00002_.flac, and so on.
• format - flac, mp3, or opus.
• quality (optional) - bitrate for MP3/Opus: V0, 64k, 96k, 128k, 192k, 320k. Ignored for FLAC, which is lossless by definition.

The output directory is ComfyUI's standard output/ folder, so filename_prefix acts like the filename prefix on the built-in Save Image node - same mental model, same folder-watching habit. Metadata (your prompt and the workflow) gets embedded in the file, which is a genuinely nice touch if you batch-render and need to know later what settings produced what.

Choosing a format

The rule of thumb is boring but right: keep flac while you're iterating, convert to mp3/opus only when you're actually exporting. FLAC is lossless and instant; MP3 at 320k is for when a track is done and you want to share it; Opus at 128k is the best quality-per-byte if you're generating lots of takes. One practical note: the file format and the quality choice don't affect generation at all - they're purely the write step, so you can re-save the same audio any number of times without re-running the sampler.

The waveform bit

Both this node and its sibling Preview Audio draw an interactive waveform - dark background, amplitude bars, play/pause, click-to-seek, and a time readout. It's a small thing, but for music it matters more than it sounds: you can hear a render's first ten seconds and scrub to a problem section without leaving ComfyUI. The waveform is computed from the audio buffer, so it reflects exactly what's being saved.

Where it fits

It's the terminal node of a generation workflow: Model Loader → TextEncode → Generate → Save Audio. Because Generate also passes through model, vae, and the conditionings, you can branch one render into several Save nodes with different formats and prefixes - one for a lossless master, one for an MP3 draft - without re-sampling.

Nothing here needs special handling, and there's no real troubleshooting to speak of. If a save fails, it's almost always a permissions issue on ComfyUI/output/ or a missing parent folder from a prefix like audio/my-takes. Install the pack once via ComfyUI Manager (search "AceStep") or git clone https://github.com/jeankassio/ComfyUI-AceStep_SFT.git into custom_nodes/, and this node appears alongside the rest under audio/AceStep SFT.