LatentUpscale_motorway_edition

LatentUpscale_motorway_edition is the "to an exact size" cousin of LatentUpscaleBy: instead of a factor, you give it a target width and height and it resamples the latent in your Motorway pipe to match. Same core node, same logic, pipe-flavored.

Use the exact-size version when the destination matters more than the ratio. The canonical case is the hi-res pass in an img2img workflow: you've got a latent and you need it to be precisely 1024×1024 (or whatever your checkpoint's happy resolution is) before it hits the sampler. The factor version is better when you just want "1.5× bigger and let the math sort out the exact pixels." The exact-size version is better when you have a hard target - like matching another branch of your graph, or keeping dimensions aligned to a specific multiple.

One mechanical detail worth remembering: this works in the 8×-downsampled latent space, so the width/height you type are image pixels, and the node converts them internally. Latent upscaling adds no new detail by itself - it resamples what exists and gives the next sampling pass a bigger canvas to fill. The "more detail" argument is the sampler's job, and the KB's upscaling writeup is blunt that generative detail-filling is a different job from resolution-filling entirely.

The wrapper

• MOTORWAY 🚌💨 - the pipe. Required.
• INPUT_samples_key (default samples) - where the latent lives.
• upscale_method - enum: nearest-exact, bilinear, area, bicubic, bislerp. bislerp is the community default for latent work.
• width, height (INT, 0–16384, step 8, default 512) - target pixel size. Step 8 keeps things VAE-friendly.
• crop - enum: disabled or center. If the aspect ratio doesn't match the source, center crops to fit; disabled leaves the mismatch.
• OUTPUT_LATENT_key (default LATENT) - where the result lands.

Defaults read samples and write LATENT; the output port is the pipe, so pull the result with a Motorway AxB ramp (key LATENT) to feed a regular node.

Installing

Part of the Motorway pack:

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

Or ComfyUI Manager → search ComfyUI_agilly1989_motorway → Install. No models, no extra pip dependencies.

Gotchas

• Start the graph with MotorwayStart (or a primitive) so there's a pipe.
• Keys are exact strings; Key: 'samples' doesn't exist in motorway means an upstream node stored the latent under a different key.
• Beta: author's README says "CURRENTLY IN ACTIVE DEVELOPMENT (BETA)! IF THINGS BREAK ITS BECAUSE I BROKE IT," and the generated clone nodes have been dropped from releases during refactors. If it's not in the menu after install, check the pack's version.

If you've got a specific resolution in mind - a checkpoint's native size, a target for a second pass, a number your video workflow demands - this is the one to reach for. Factor math is for laziness; exact targets are for correctness.