RC Image Scale

Scaling is the operation every compositing workflow does a dozen times, and most people get it wrong by default: they stretch pixels with whatever resampler ComfyUI happened to give them and wonder why their overlaid logo looks mushy. RC Image Scale is the pack's take on doing it right - six scaling modes and four resampling algorithms, with LANCZOS as the default instead of an afterthought.

It's a utility node from the RC Image Compositor pack, and it's the quiet workhorse you reach for when you need to size an overlay to match a background before you composite it.

How it works

Two decisions, independent of each other: how to compute the target size, and how to resample when you get there.

Scaling modes (scale_method):

• percentage - multiply by scale_factor (0.1–10×).
• dimensions - exact width/height, with keep_aspect_ratio (default on) protecting you from distortion.
• fit_width / fit_height / fit_longest / fit_shortest - the "fit into this frame" family. Fit a 1024×1024 image to width 512 and it becomes 512×512; fit a 2048×1024 image and you get 1024×512. These are the ones that keep your aspect ratio while matching a target edge.

Resampling (resampling):

• LANCZOS - default. Highest quality, and on modern hardware the "slow" penalty is a rounding error. Just use it.
• BICUBIC - high quality, slightly softer.
• BILINEAR - medium. Fine for downscaling thumbs.
• NEAREST - pixel art. Only correct choice when you're scaling something that must stay pixel-sharp, like a sprite or a map.

The inputs that matter

• image - input tensor.
• scale_method and scale_factor (or width/height, depending on mode).
• resampling - LANCZOS unless you have a reason not to.
• keep_aspect_ratio - leave it on.

Output is a single IMAGE tensor at the new size.

How to install it

Ships with the RC Image Compositor pack:

cd ComfyUI/custom_nodes
git clone https://github.com/kj863257/ComfyUI_RC_Image_Compositor
cd ComfyUI_RC_Image_Compositor
pip install -r requirements.txt

Restart ComfyUI, or search "RC Image Compositor" in ComfyUI Manager. Dependencies are just pillow, numpy, and opencv-python.

Common issues

Two recurring mistakes. First, dimensions mode with keep_aspect_ratio off will happily squish a 1024×1024 into 512×1024 - there are legit uses (matte pulls, aspect-stretch VFX plates) but if your overlay suddenly looks wrong, this is the first thing to check.

Second, upscaling by a large factor here is not the same as real upscaling. LANCZOS interpolation up to 2× or 4× looks fine for compositing, but if you're trying to genuinely restore detail, this isn't the node - that's a job for a dedicated upscaler model. Use Image Scale to fit, not to enhance.

And a workflow note: the pack's RC Image Compositor has its own scale_mode that scales the overlay relative to the background while positioning it. If your only goal is "put this logo on that image at 30% of its width," you can skip the separate scale step entirely.