Ksampler Parameters Default JK๐
One knob for steps, CFG and denoise across your whole graph
- STEPS
- CFG
- DENOISE
This is a "parameters default" node: a place to park your sampling numbers so the rest of the graph reads them from one spot instead of you editing the same steps value in four different KSamplers. Ksampler Parameters Default JK takes steps, cfg, and denoise, and hands them straight back out as typed outputs. It doesn't sample anything. It doesn't even validate. It's a shared constant, with a UI.
How it works
Mechanically there's nothing to it - the node returns exactly the three values you type in, with sane ranges enforced on the widgets:
steps- integer, default 20, range 1โ10000.cfg- float, default 8.0, range 0โ100, step 0.05.denoise- float, default 1.0, range 0โ1, step 0.01.
The point is structural. In JakeUpgrade's group-node workflows, one parameters node feeds several samplers - a main pass and a refine pass, say - so bumping steps once re-syncs the whole graph. If you've ever changed denoise in the refine stage and then discovered the base stage was still at the old value, you know why this exists.
Inputs and outputs
stepsโSTEPS(INT)cfgโCFG(FLOAT)denoiseโDENOISE(FLOAT)
Wire STEPS/CFG/DENOISE into any KSampler, KSampler Adv, or JakeUpgrade group node that exposes matching inputs. For the plain KSampler you'll usually hardcode seed and pick the sampler/scheduler yourself; this node covers the numbers.
Installing it
Ships in ComfyUI-JakeUpgrade:
- ComfyUI Manager โ search "JakeUpgrade" โ install โ restart.
- Or:
cd ComfyUI/custom_nodes
git clone https://github.com/jakechai/ComfyUI-JakeUpgrade
cd ComfyUI-JakeUpgrade
pip install -r requirements.txt
Windows portable: run install.bat or ../../../python_embeded/python.exe -s -m pip install -r requirements.txt.
Common issues
- The default
cfgof 8.0 is an SDXL-era default. On modern flow-matching models - Flux, Wan, and friends - "CFG 8" is wrong, and guidance works differently or not at all. This node is a dumb pipe, so you have to know what number your model actually wants. Don't treat 8.0 as a recommendation, treat it as a starting guess from 2023. - It duplicates ComfyUI's own primitives. Recent ComfyUI has
Primitive/Anythingnodes that can feed KSampler inputs directly, and you can fake a shared constant with a group input. This node is nicer when you're already in the JakeUpgrade ecosystem and want typed, labeled outputs instead of generic ones. - Nothing prevents a bad combination.
cfg=0withdenoise=1will run and produce mush. The node won't stop you; it's a convenience, not a guardrail.
If your workflow has one sampler, skip it. If it has three and you keep editing them out of sync, this is the node that makes the graph feel sane again.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| steps | INT | 201โ10000 | Number of sampling steps |
| cfg | FLOAT | 8.000โ100 | Classifier-Free Guidance scale |
| denoise | FLOAT | 1.000โ1 | Denoising strength (1.0 = full denoise) |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| STEPS | INT | โ |
| CFG | FLOAT | โ |
| DENOISE | FLOAT | โ |