StableCascade_StageB_Conditioning64

If you're still running Stable Cascade, you already know it's not one model pretending to be simple - it's a three-stage pipeline (Würstchen under the hood) where a tiny Stage C latent gets expanded by Stage B before Stage A turns it into pixels. StableCascade_StageB_Conditioning64 is the plumbing that makes Stage B actually pay attention to what Stage C produced, instead of generating from your text prompt alone and hoping it lines up.

Cascade is a bit of an orphan model at this point - it had genuinely strong prompt adherence and aesthetic scores when it launched, but SD3's announcement landed while the ecosystem was still deciding whether to invest, and by the time SD3 disappointed everyone had already moved on to waiting for Flux. Most packs quietly dropped Cascade support. RES4LYF didn't, partly because its author also built UltraCascade on top of it and clearly never lost interest - this node is part of that leftover, still-maintained plumbing.

How it works

Stage C isn't just "step one of two." Its latent output is meant to condition Stage B, not just get decoded past it - Stage B needs to know what Stage C actually generated to expand it coherently rather than inventing its own interpretation from the text alone. This node takes whatever conditioning you're already sending to your Stage B sampler and folds Stage C's sampled latent into it, handing back an updated conditioning that carries both. There's nothing to tune here; it's a single-purpose wiring node, not a control panel.

The inputs and outputs that matter

• conditioning (CONDITIONING) - whatever you were already going to feed Stage B, typically your CLIP text encode for the positive prompt.
• stage_c (LATENT) - the sampled output straight off your Stage C sampler, before any Stage B decoding happens.
• Output: conditioning - wire this directly into your Stage B sampler's positive input in place of your raw CLIP conditioning.

How to install it

Via ComfyUI Manager: search RES4LYF, install, restart.

Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/ClownsharkBatwing/RES4LYF
cd RES4LYF
pip install -r requirements.txt

If you're on a portable install, point at the embedded Python's pip.exe instead of a venv one. The pack's newer sampler-submenu UI wants rgthree-comfy installed alongside it too - not required for this specific node, but worth having regardless if you're setting RES4LYF up fresh.

Common issues & troubleshooting

Skipping this node entirely. It's tempting to just wire your CLIP conditioning straight into Stage B and call it done - you'll still get an image, but Stage B won't actually know what Stage C built, and results come out noticeably muddier and less coherent than workflows that use this node properly.

Wrong checkpoint stack. Stage B needs Stable Cascade's own checkpoints (stage_b, stage_c, and the effnet/Stage A pieces), not a regular SD or SDXL model. Feed it the wrong architecture and it won't error cleanly - it'll just produce garbage.

Looking for it in the newer example workflows and not finding it. RES4LYF's current documentation pushes most people toward the unified ClownsharKSampler for flow-matching models (Flux, SD3.5, Wan, and so on). This Stage B/C conditioning node is specifically for the older, two-sampler Cascade pipeline, so don't expect to see it referenced outside Cascade-specific workflows.