SamplerLCMUpscale_motorway_edition
Speed tech for distilled models, bus edition
- MOTORWAY 🚌💨
- MOTORWAY 🚌💨
SamplerLCMUpscale_motorway_edition is a clone of ComfyUI's built-in SamplerLCMUpscale, and it's the most niche sampler in this batch - if you don't already know you need it, you probably don't. It exists for one scenario: you're sampling with an LCM (Latent Consistency Model) or other step-distilled checkpoint, you want the result at a higher resolution than the latent started at, and you want that upscale to happen during sampling rather than as a separate upscale pass afterward.
The LCM story is worth a sentence because it explains why this node behaves the way it does. LCM distills a model so it can jump to the answer in 4–8 steps instead of 20–50 - the tradeoff being quality at the edges. This sampler takes that idea and adds an upscale step inside the denoising trajectory: when it hits the configured scale point, it upscales the latent mid-sampling, so the sampler keeps refining at the larger size instead of handing you an image you then have to upscale separately. It's clever, it's fast, and it's the kind of thing that only makes sense with the distilled-model toolchain around it. With a normal checkpoint you'd be better off with a plain sampler plus a dedicated upscaler.
The widgets:
- scale_ratio (default 1.0, 0.1–20) - how much the latent is upscaled during sampling. 1.0 means no upscale; 4.0 would target 4×.
- scale_steps (default -1, -1–1000) - when the upscale happens. -1 means "upscale on the last step"; a positive number upscales at that step so the rest of the sampling runs at the larger size.
- upscale_method -
bislerp,nearest-exact,bilinear,area, orbicubic. Same choices ComfyUI uses everywhere else for latent resizing; bislerp is the usual pick.
Like every node here it's a sampler factory - model in, SAMPLER out - so you still feed the result into SamplerCustom's sampler port (or a SamplerCustom_motorway_edition on the bus).
What it does on the Motorway
A generated *_motorway_edition clone from ComfyUI_agilly1989_motorway: a named-key bus where values ride under typed keys instead of wires. MotorwayStart starts a bus, Motorway AxB ramps store/fetch keys, and each motorway-edition node pulls inputs from the bus by key and writes outputs back by key, always returning the bus. Model in via the family's INPUT_…_key convention (default model), sampler out under OUTPUT_SAMPLER_key (default SAMPLER).
Installing it
Part of ComfyUI_agilly1989_motorway. ComfyUI Manager: search ComfyUI_agilly1989_motorway, install, restart. Or:
cd ComfyUI/custom_nodes
git clone https://github.com/agilly1989/ComfyUI_agilly1989_motorway
No extra dependencies, no model files.
The catch
Standard pack gotchas: release 1.1.7 ships with the clones disabled (uncomment the two ClonedNodeMapping lines in custom_nodes/ComfyUI_agilly1989_motorway/__init__.py and restart), and the README is upfront that this is a solo dev's beta. On top of that, this particular sampler is niche - the "upscale during sampling" trick only pays off with LCM-style distilled models, and even then it's one of those nodes people try, find they can't match their normal upscale pipeline's quality, and quietly uninstall. Know what it's for before you reach for it.
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| scale_ratio | FLOAT | 1.000.1–20 | — |
| scale_steps | INT | -1-1–1000 | — |
| upscale_method | COMBO | 5 options: bislerp, nearest-exact, bilinear, area, bicubic | |
| OUTPUT_SAMPLER_key | STRING | SAMPLER | — |
| MOTORWAY 🚌💨opt | MOTORWAY 🚌💨 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| MOTORWAY 🚌💨 | MOTORWAY 🚌💨 | — |