RL Set Upscale Data

RL Set Upscale Data is the smallest node in the pack and doesn't pretend to be more: it's width and height, passed straight through. Its entire job is to surface the scaling options you set inside Reallusion's AI Render plugin as plain integers your workflow can read. Think of it as a named constant on the graph - the plugin writes the target size, this node carries it, and whatever upscale node you've wired downstream consumes it.

That's it. There's no upscaler in here, no model, no algorithm. Read the source and process() literally returns the two integers it received. If you're expecting this to be a Reallusion-branded SUPIR or a 4x ESRGAN runner, this isn't that - and the KB's upscaling essay explains why the distinction matters: "bigger" and "better detail" are different jobs, and this node only knows the first one.

What you actually set

• width / height - target dimensions, 64 to 8192, defaulting to 512. The node trusts you, so the real question is what your downstream upscaler can chew: 8192 is a lot of VRAM if your pipeline tries to do it in one pass, and classic 4x ESRGAN-family models expect multiples of four, so a size like 513 or 514 from a plugin default can come out crooked.

Outputs are the same width and height as INTs. Wire them into a Empty Latent Image, an upscale node, or whatever sizes your final output. Because the plugin controls the value, you can treat the node as read-only in normal use - it exists to carry the plugin's decision, not to override it.

Install

It's part of the four-node Reallusion pack, which has no dependencies at all:

cd ComfyUI/custom_nodes
git clone https://github.com/reallusion/ComfyUI-Reallusion

Restart ComfyUI and it appears under Reallusion (or grab it from ComfyUI Manager by searching "Reallusion"). No requirements.txt, no model downloads - you still supply your own upscale model in the part of the graph that does the actual resizing.

The honest summary

If you're building an AI Render workflow by hand, this is the node you'll barely notice: it sits there holding two integers so the rest of the graph knows the target size. If you were hoping it was the upscaler itself, that's the one thing to unlearn - the upscale model lives elsewhere, and this node just tells it where to aim. Keep the width and height sensible for your GPU, remember the "multiples of four" rule for ESRGAN-style models, and this node will never be the thing that breaks.