CondLinearInterpolate

The bracket scheduling syntax this pack is best known for - [cat:dog:0.5] - is a hard switch: cat conditioning up to the breakpoint, dog conditioning after it. CondLinearInterpolate is a different, experimental answer to the same general problem - going from one prompt's conditioning to another's - but instead of a hard cutover, it steps between them incrementally, blending the two.

What it does

You give it two already-encoded conditionings, a start and an end, and it linearly interpolates between them - walking from start's conditioning toward end's in increments, rather than switching all at once. That produces a smoother morph across the denoising range instead of a visible seam at one breakpoint, which is the kind of thing worth reaching for when a hard [a:b:x] schedule switch is producing an obvious "before" and "after" look instead of a gradual transition.

Living under promptcontrol/exp tells you where this sits in the pack's maturity - it's an experimental technique, not part of the documented core-node list (PCLazyTextEncode, PCLazyLoraLoader, PCTextEncode) the README walks through. Treat it as something to try and evaluate against your specific use case rather than a guaranteed-solid tool.

Inputs and outputs

Required: start and end, both CONDITIONING - the two prompts (already encoded, from PCTextEncode or any standard text-encode node) you want to blend between. Optional: until (float, 0–1, default 1) - how far through the range the interpolation runs; step (float, 0–1, default 0.1) - the increment size between interpolation points. Output: CONDITIONING.

Installing it

Ships with the rest of the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/asagi4/comfyui-prompt-control

Or ComfyUI Manager, search "ComfyUI Prompt Control". No model downloads.

Where people get tripped up

Because this is experimental and undocumented in the README's own walkthrough, don't assume its output shape or exact interpolation behavior matches whatever you'd get from bracket-scheduling the same two prompts through PCLazyTextEncode - they're different mechanisms (discrete graph-of-timestep-ranges versus incremental conditioning interpolation) and will produce different results even aiming at a similar effect. If you want the well-trodden, documented path for transitioning between prompts over time, start with [promptA:promptB:0.5]-style scheduling in PCLazyTextEncode first; reach for CondLinearInterpolate specifically when you've tried that and want to test whether a smoother blend changes the seam you're seeing, and compare the two side by side on a fixed seed rather than assuming one is strictly better.

The step value is also worth watching for cost: a smaller step means more interpolation points computed between start and end, which is more work for a smoother gradient. There's no documented guidance on the right value for a given use case - treat 0.1 as a reasonable starting point and adjust by eye.