KSampler Info (No CFG)

Somewhere around when Lightning and Turbo models showed up, "just leave CFG at 7" stopped being safe advice. Distilled models bake their guidance in during training, so cranking CFG on top of them oversaturates and artifacts - the standard advice is CFG at 1.0 or even 0. This node is the version of Sage Utils' "KSampler Info" for exactly those models: it collects the sampler settings but hard-codes cfg to 1.0, so there's no CFG field to accidentally bump.

How it works

Same shape as the regular Sage_SamplerInfo: you set seed, steps, sampler_name, and scheduler, and the node bundles them into a single SAMPLER_INFO object. There's no cfg input at all - that's the entire difference. The optional advanced_info input still works the same way, merging extra advanced sampler settings into the bundle if you provide one.

You route the output exactly like its sibling: into a Sage "KSampler w/ Sampler Info" node (or the tiled/audio variants) to run the generation, and into a Construct Metadata node so the settings land in your image metadata. Because CFG is fixed at 1.0 inside the node, the metadata will always record what you actually ran - no more remembering whether this workflow was the CFG-7 one or the CFG-1 one.

When to reach for it

If your checkpoint is a distilled/Turbo/Lightning-style model - the kind where the recommendation is "1-4 steps, CFG at 1 or lower" - this is the right pick. It's also the right pick for models that are straight-up guidance-free (some video models, for instance, where CFG is effectively disabled). If you're running a regular SDXL/Pony/Illustrious checkpoint that lives at CFG 5-7, use the regular Sage_SamplerInfo instead - this node's whole reason to exist is removing a knob you shouldn't touch.

The other, subtler advantage: it keeps your sampler controls honest. When the CFG field is sitting right there on the node, people bump it "just a little" and then wonder why their distilled model got crunchy. Remove the field and the temptation goes with it.

Installing it

Same pack, same install. ComfyUI Manager (search Sage Utils) or:

cd ComfyUI/custom_nodes
git clone https://github.com/arcum42/ComfyUI_SageUtils
cd ComfyUI_SageUtils
pip install -r requirements.txt

Restart and you're set. The pack's only pip dependency is dynamicprompts; nothing here needs model files.

The trap to avoid

Don't grab this node out of habit for a normal CFG model - with cfg locked at 1.0 you're effectively running with no classifier-free guidance, and prompt adherence will visibly suffer. It's not broken, it's just the wrong tool. And like the regular Sampler Info node, remember it only plugs into Sage's own KSampler nodes, not a stock ComfyUI KSampler.