Clocked Sweep Values

Clocked Sweep Values takes a clock list and a values list, and spreads those values evenly across the clock. The output length always matches the clock length - that's the whole contract, and it's more useful than it sounds.

The use case is sweeping a float parameter across the length of a run. Say you want denoise to start high and taper off over a 20-step sequence. You feed it a clock of step indices and the two endpoint values, and it hands back a list where each value is repeated for roughly clock_length / values_count slots, with any remainder spread across the first few values. Wire that list into whatever accepts per-step floats and you've got a ramp without writing a single loop.

How the sweep works

The parsing is pleasantly forgiving. Both clock and values accept a JSON list ([0.9, 0.3]), comma or space-separated numbers, a single integer string like 20 (interpreted as 1..20), or an actual list/tensor input. If you leave clock blank, its length is inferred from values, which is handy when all you care about is the sweep. The output is a FLOAT list named values.

Inputs and output

Two inputs, one output:

• clock - defines the output length. Step indices or just "20".
• values - the numbers you're sweeping across.

Gotchas

The main trap: values can't have more entries than clock - the node raises a clear error rather than guessing. Also note it does not interpolate between values; each value is repeated in blocks. If you want a smooth linear ramp, give it more values rather than two endpoints and expecting a gradient.

Installing

This node ships in the ttulttul/ComfyUI-Taylor-Attention pack alongside its batch and prompt-list siblings, and the install path is shared. Use ComfyUI Manager (search "Taylor-Attention") or:

cd ComfyUI/custom_nodes
git clone https://github.com/ttulttul/ComfyUI-Taylor-Attention

Restart ComfyUI, then install the pack's dependencies - it's a research pack with real requirements (torch, einops, pyiqa, comet-ml, dreamsim and friends) and it needs a recent ComfyUI with the v3 node API:

uv pip install -e custom_nodes/ComfyUI-Taylor-Attention

Ignore the README's stale folder name (ComfyUI-Approximate-Attention) - point uv at the folder you actually cloned. No model downloads; the checkpoints this pack uses live under ComfyUI/models/approximate_attention/.

Everything in this pack is marked experimental, and ClockedSweepValues is no exception. It's a small tool, but once you're running parameter sweeps by the dozen, having the distribution math done for you is quietly worth the install.