StableCascade_SuperResolutionControlnet_motorway_edition

Stable Cascade's official super-resolution workflow works like a ControlNet: you take a low-res image, encode it, and use it as guidance while stage B samples at higher resolution, instead of just blurring an upscale. StableCascade_SuperResolutionControlnet is the node that builds that control input. This is the Motorway edition, so the image and VAE come out of the pipe by key.

Fair warning before you go deep: in core ComfyUI this node lives under the _for_testing category, and it's been there a long time. Stable Cascade itself never really caught on the way SDXL or Flux did, and the super-res path is the fiddliest corner of a niche architecture. It works - you feed an image and a VAE in, and you get back the controlnet input plus both stage latents - but you should expect rough edges and sparse community help when something's off.

What to set

• INPUT_image_key (default image) - the Motorway key holding the low-res IMAGE to upscale from.
• INPUT_vae_key (default vae) - the Cascade VAE.
• OUTPUT_controlnet_input_key (default controlnet_input) - the control input for the stage B sampler.
• OUTPUT_stage_c_key (default stage_c) - the stage C latent.
• OUTPUT_stage_b_key (default stage_b) - the stage B latent.

Output is the MOTORWAY with all three stored. The controlnet_input gets wired into a ControlNet apply against your stage B conditioning, and the latents feed the samplers.

Installing it

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

Restart; it's under agilly1989 Nodes → Motorway-ed → _for_testing → stable_cascade. Manager installs the pack too. No pip dependencies - this is a pure wrapper over ComfyUI's own node.

Where people get burned

Everything that bites elsewhere in Cascade bites here, plus one more. The VAE and compression must match your checkpoint, or the encode silently produces garbage latents. And because this is _for_testing-class tech on top of a beta pack, error messages when things go wrong can be cryptic - the pack's own catch-all ('MotorwayClass' object has no attribute 'hash_') just means a key you referenced isn't in the pipe, so double-check the keys writing image and vae.

Honest take: unless you're already committed to Stable Cascade, this node is not the reason to install the pack. If you just want better upscaling, a standard latent upscale or an external upscaler will give you a better time than debugging Cascade super-res. If you're here because you are running Cascade, the motorway-ed version just saves you from dragging three output wires.