Audio Modulation

Want your generated music to stop sounding like a dry export and start sounding processed? Audio Modulation is the stage you grab. It's a one-node guitar-pedal rack with six effects - phaser, flanger, chorus, vibrato, tremolo, and pulsator - selected via the mode combo. There's no AI anywhere in here; it's real DSP running on an FFmpeg filter graph, so it's instant, deterministic, and you can stack a whole pedalboard of them without worrying about VRAM.

The honest read: it's not six separate effects, it's six dial sets sharing one node, and only the dials for your chosen mode apply. The ph_ params drive the phaser (delay 0–5ms, decay, LFO speed, triangular vs sinusoidal sweep), the fl_ params drive the flanger (delay, depth, regeneration, width, speed, phase), and chorus_preset swaps between single, double, and triple chorus thickness. Vibrato and tremolo just take an LFO frequency and depth (lfo_f, lfo_d), and pulsator is the trance-gate rhythm effect with rate (pu_hz), pu_amount, pu_width, and an LFO waveform (pu_mode). Set the mode, touch the dials for that mode, leave the rest alone - they're inert when not selected.

Mechanically, each mode maps to a stock FFmpeg filter: aphaser, flanger, chorus, vibrato, tremolo, apulsator. The node builds the filter spec from your widgets and runs it through PyAV. That's why it's fast and why it has no model files. A nice side effect: the stage also emits an fx_spec output alongside audio - that's the metadata ComfyTV's FX Chain uses to stack this effect with others and render them in a single pass. If you're chaining five modulation stages, wire their fx_specs together instead of running five separate renders.

Like every AudioFX stage, the input is audio or video - drop a clip in and it processes the clip's audio track. The three hidden inputs (force_run_token, project_id, parent_output_id) are ComfyTV's per-node Run plumbing; don't touch them.

Where people get burned: the flanger's fl_regen goes negative-to-positive (−95 to +95) and positive values push into self-oscillation territory, which can blow past 0 dB and clip hard. If your output suddenly sounds like a howling modem, check your peak levels after a flanger with high regen, and don't forget that this node outputs raw processed audio - there's no limiter riding shotgun. Keep the phase, start subtle.

Install is the pack-wide story - ComfyUI Manager search "ComfyTV", or git clone https://github.com/jtydhr88/ComfyTV into custom_nodes/, then a full backend restart. On Desktop/macOS with multiple ComfyUI installs, clone into the running instance's absolute path. No dependencies beyond what ComfyUI already ships, no models, no GPU needed.