ControlFoley Advanced Generate

ControlFoley Advanced Generate is the same generation engine as the plain ControlFoley Generate - identical inputs, identical mode mapping, identical flow-matching guts - with three additions bolted on for people building reusable chains: an enabled switch, a silent_audio_on_error fail-safe, and a status string output. If you're just trying to make one clip sound right, use the regular node. If you're building a workflow you intend to run a hundred times, this is the one.

What the three extras actually do

• enabled - flip it off and the node skips generation entirely and returns a correctly-sized silent audio buffer instead of running the model. That's enormously useful when you're iterating: you can mute a branch of the graph without deleting nodes and rewiring, which keeps the rest of the pipeline alive while you work on something else.
• silent_audio_on_error - by default a failed generation throws and fails the run, the normal ComfyUI behavior. Turn this on and an error inside the generator gets swallowed and replaced with silence plus a status string that tells you what failed. Great for long unattended batch runs where one bad clip shouldn't kill the queue.
• status - a fifth output (STRING) that reports what happened: "Generation completed successfully.", "Generation disabled; returned silence.", or, in silent-error mode, the exception message. Wire it into a text display or just glance at it in the UI.

Everything else matches ControlFoley Generate: connect video / video_input / images (one of the three) to pick V2A, TV2A/TC-V2A, AC-V2A, or pure T2A; reference_audio_path for the audio-controlled mode; num_inference_steps is the string fixed (25) or an integer 1–100; guidance_scale defaults to 4.5. The same caveats apply - staged_offload defaults to true but is ignored with a console note on the public upstream source, and the first run downloads roughly 16GB of weights. (The low_vram flag, set on the Model Loader rather than here, is a text-only path; this node hard-errors if a video or reference audio is connected while it's on.)

Where it fits in the chain

The intended shape (see the bundled 06_advanced_chain.json example) is the full pipeline: Dependencies Loader → Model Loader → optionally Torch Compile → this node → Save Audio. Because the advanced generator carries its own status, it's the natural final consumer in a chain you want to monitor or auto-batch. It also shares the exact same generate_audio implementation as the base node - the source literally calls into it - so output quality is byte-identical; you're paying for the controls, not a different model.

Outputs

audio (AUDIO) goes to Save ControlFoley Audio or the Video-Audio Muxer; sample_rate (INT), inference_time_sec and peak_vram_gb (FLOATs) are your instrumentation; status is the new conversation partner. Install the pack via ComfyUI Manager ("ControlFoley Official") or by cloning YJX-Research/comfyui-controlfoley-official into custom_nodes and running pip install -r requirements.txt - install only missing packages one at a time, since that requirements file holds version ranges that can upgrade shared deps like transformers or numpy out from under your existing ComfyUI.