Ksampler Adv Parameters Default JK🐉

If you've opened one of Jake's Wan 2.2 workflows and wondered what "High+Low" means, this node is the answer hiding in plain sight. Ksampler Adv Parameters Default JK packages the step schedule for a two-phase sampling trick: sample with high CFG early, then drop to a low CFG for the rest. The pack calls it KSampler (High+Low), and this little node is what feeds the group node its numbers.

What the High+Low trick is

Video models - Wan 2.2 especially - respond well to starting the denoise with a strong guidance signal (high CFG) to lock in structure and composition, then switching to a gentler CFG for the remaining steps so you don't get that oversaturated, crispy look. The switch has to happen at a specific step, which is exactly the one thing you control here.

The mechanism (and it's just math)

This node takes one real input - switch_at_step (default 10) - and returns a complete step schedule:

• start_at_step - always 0.
• switch_at_step - your value. The step where the sampler swaps settings.
• end_at_step - 10000, i.e. "run to the end."
• enable / disable - the strings "enable" and "disable" that the group node uses as a switch signal for the two phases.

So with switch_at_step = 10, you get "high CFG from 0 to 10, low CFG from 10 to the end." Change the one number and the whole schedule follows. It's a "default" node in the sense that it gives the group node sane defaults - you're not expected to micro-manage all five outputs, just the step where the handoff happens.

Wiring it up

You almost never build this from scratch. It ships in the pack's 🐉 JK:::KSampler (High+Low) group nodes (v2.0.0 added both), and those group nodes expose the switch step as a surfaced input. If you're hand-building the same pattern, feed the five outputs to wherever your custom sampler setup consumes start/switch/end step ranges and the enable/disable toggles.

Install

Clone the pack, install requirements, restart:

cd ComfyUI/custom_nodes
git clone https://github.com/jakechai/ComfyUI-JakeUpgrade
pip install -r requirements.txt

No models involved. One honest note: end_at_step = 10000 is effectively "infinity" - that's deliberate, so the schedule always runs to whatever the sampler's real step count is. Don't "fix" it to match your step count, or you'll truncate generations.