MD: LFO Generator (Automator)

An LFO - low-frequency oscillator - is the musician's trick of wiggling a parameter on a slow, repeating wave. MD_LFO_Generator brings that to ComfyUI: it produces a list of values that oscillate between two bounds over a fixed number of steps, so you can sweep a denoise level, a strength, a scale - any FLOAT input - over the course of a batch or an animation instead of holding it constant.

The neat part is the dual output. Most ComfyUI modulation nodes make you pick one style: a value list to splice into a batch, or a single value for the current frame. This one gives you both. And it renders a plot, because guessing whether your wave actually looks like what you wanted is the fastest way to waste a render.

How it works

The core is a tiny waveform generator. Pick a shape - sine, triangle, sawtooth, square, pulse, noise, or the easing variants - and it samples that wave across steps frames, mapped so the wave's low point hits amplitude_min and its high point hits amplitude_max. frequency sets how many full cycles fit across the whole run (not per-second, which trips people up - 1.0 means one cycle over the entire list). phase_offset shifts where the wave starts, so 0.5 begins halfway through a cycle. The Random waveform is seeded, and noise_smoothing runs a convolution over it so random jumps don't make your parameter spasm frame to frame.

The inputs that matter

• waveform - 12 shapes. Sine is the sane default; sawtooth for a ramp; square for hard on/off.
• frequency - cycles across the whole duration.
• amplitude_min / amplitude_max - the floor and ceiling of the sweep. These are the two you'll touch constantly.
• steps - how many values the list holds. Match it to your batch size or total frames.
• phase_offset - where the wave begins.
• seed / noise_smoothing - only matter for the Random waveform.

Outputs: current_value (a single FLOAT, ideal when wired with the optional current_step input so it tracks a changing frame index), value_list (the full FLOAT list for batch/broadcast workflows), list_as_string (for text or debugging), and lfo_plot (an IMAGE preview of the wave).

Installing

Ships with MD Nodes:

cd path/to/ComfyUI/custom_nodes
git clone https://github.com/MDMAchine/ComfyUI_MD_Nodes.git
cd ComfyUI_MD_Nodes
pip install -r requirements.txt

Or ComfyUI Manager → MD Nodes → install, then restart. The plot output needs matplotlib; the values themselves don't.

Common issues

The most common mistake is reading frequency as "cycles per second" - it's cycles per full run, so a frequency of 1 over 20 steps is a single gentle swell, not a wobble. And remember that a plain LFO output doesn't change anything by itself: it's only useful connected to a parameter that varies per frame or per batch item. If your workflow expects one fixed number, wiring current_value with a fixed current_step just hands you back a constant - pick the output that matches how your consumer node consumes it.