EasyCascadeKsampler

Stable Cascade had one of the odder trajectories of any model in this ecosystem's history: real hype at launch in February 2024 ("Stable Cascade is out!" pulled over 500 upvotes), genuinely strong prompt adherence for the time, and then it got squeezed to death between two other announcements - SD3 got announced while people were still figuring out Cascade's workflows, and by the time SD3 disappointed, Flux had already eaten the room. Nobody really went back. If you're picking an architecture fresh today, this isn't the one. If you've got existing Cascade work, or you're specifically curious about its three-stage Würstchen pipeline, this node - the simple version of its stage-C sampler - is how Easy-Use runs it.

Why Cascade needs a different sampler node at all

Stable Cascade isn't a single-latent pipeline like SD or SDXL. It runs in stages: stage C works in an aggressively compressed latent space (which is the source of its speed and low VRAM claims), and later stages decompress that down toward pixels. This node handles the stage-C pass - the "simple" one, meaning it leans on settings already configured in your pipe rather than exposing every sampling parameter directly, which is the fullCascadeKSampler node's job instead.

The inputs and outputs that matter

pipe in, required, carrying your loaded Cascade model and conditioning. image_output is a shared Easy-Use convention across its sampler nodes - Hide, Preview, Save, Hide&Save, Sender, or Sender&Save - controlling what happens to the result without needing a separate Preview or Save node bolted on afterward. link_id is for Easy-Use's internal sender/receiver linking when you're using Sender modes to pass images between disconnected parts of a large graph. save_prefix is your filename prefix when saving.

The one optional input, model_c, lets you override the stage-C model carried in the pipe - useful if you're experimenting with a different Cascade checkpoint without rebuilding your whole loader chain.

Two outputs: pipe (carrying forward whatever a later stage needs) and image - and this node is flagged as an output node, meaning it can display or save results directly depending on your image_output setting.

Installing it

ComfyUI Manager: search ComfyUI Easy Use, install, restart. By hand:

cd ComfyUI/custom_nodes
git clone https://github.com/yolain/ComfyUI-Easy-Use

Windows: double-click install.bat. Linux/Mac: run pip install -r requirements.txt yourself. Restart ComfyUI. Stable Cascade's own checkpoints (stage B and stage C separately) are a meaningful download on top of the node pack itself - budget for that before you start.

Where people get stuck

The most common source of confusion isn't this node, it's Cascade's own two-stage structure - if your output looks wrong or incomplete, check that you're actually feeding a properly loaded Cascade pipe with both stage models available where the pipeline expects them, not just stage C. Second, image_output set to Hide is a frequent "why isn't anything showing up" moment - it's working correctly, it's just not previewing; switch to Preview or Save if you want to actually see the result. And given how little active Cascade tooling development is happening anywhere in the ecosystem now, if you hit a genuinely obscure error, it's worth checking whether it's a known Cascade quirk from 2024 rather than assuming it's something new - the model's issues are, at this point, mostly already documented somewhere from when people were actually using it seriously.