StableCascade_StageB_Conditioning_motorway_edition

In a Stable Cascade workflow, stage C and stage B speak different languages. Stage C produces a heavily compressed latent; stage B needs that latent plus your text conditioning, fused into a conditioning object it can actually sample from. StableCascade_StageB_Conditioning is the bridge between the two stages. This is the Motorway version, so both inputs are pulled out of the pipe by key.

If you've been building Cascade workflows and just copied the node placement without understanding it, here's the mechanism: stage B doesn't take the raw stage C latent and the raw conditioning separately - it needs them combined, because during stage B sampling the model is guided both by your prompt and by the structure already baked into that stage C latent. This node does that fusion. Get the wiring wrong and stage B either ignores your prompt or ignores the composition entirely, which shows up as stage B output that has nothing to do with what stage C just drew.

What to set

• INPUT_conditioning_key (default conditioning) - the Motorway key holding the CONDITIONING from your CLIP text encode (plus any LoRA/ControlNet-modded conditioning).
• INPUT_stage_c_key (default stage_c) - the Motorway key holding the stage C latent you sampled in the first stage.
• OUTPUT_CONDITIONING_key (default CONDITIONING) - where the fused stage-B conditioning is written.

Output is the MOTORWAY with the fused CONDITIONING stored. Feed that into the stage B sampler (with stage_b from an empty-latent or VAE-encode node) and you're set.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/agilly1989/ComfyUI_agilly1989_motorway

Restart; it's under agilly1989 Nodes → Motorway-ed → conditioning → stable_cascade. Manager installs the pack too. No extra Python dependencies - a pure wrapper around ComfyUI's own node.

Where people get burned

Order of operations. The stage C sampler has to actually run before this node executes, or the stage_c key won't be in the Motorway yet - and you'll get the pack's signature 'MotorwayClass' object has no attribute 'hash_' error, which just means "key not found." Check that anything writing stage_c is upstream in execution order, not just placed earlier on the canvas.

Second, the conditioning has to be Cascade-compatible. If you ran your text through a standard SD/SDXL CLIP encode rather than the Cascade text encoder, stage B will be confused. Match the model family. And the usual pack warning: it's an active beta, and the auto-cloned motorway-ed nodes have been enabled and disabled across releases - a disappearing node after an update is upstream churn, not a broken install.