Scale

IsekaiScale is the pack's resizer: one node that can scale an image by a factor, to exact dimensions, or by a percentage, with four resampling methods to pick from. If you already know ComfyUI's core image scaling, this is a more flexible spin on it in a single node - though "more flexible" comes with a mode you can absolutely misread on the first try.

How it works. It converts to PIL and calls resize with dimensions computed from whichever scale_method you pick:

• Factor - multiply by scale_x / scale_y. Factor 2.0 doubles the size, 0.5 halves it. Straightforward.
• Dimensions - resize to exact width / height (defaults 512×512, range 1–8192). What you ask for is what you get, aspect ratio be damned - a 1024×1024 image scaled to "Dimensions 512×512" becomes a 512×512 image, full stop.
• Percentage - here's the trap: scale_x is treated as a percentage, so 50 means half, not 50×. The default scale_x/scale_y of 1.0 under this mode means 1% - an image scaled to 1% of its size. It's consistent with how the node labels it, but if you flip modes and leave the sliders alone, "Percentage" at the default quietly nukes your image into a thumbnail. Set 100 for "no change."

resampling gives you Nearest, Bilinear, Bicubic, or Lanczos. For downscaling, Lanczos is the usual quality pick; Nearest is the right call when you're deliberately doing pixel-art or hard-sampling work.

Output is a single image (IMAGE tensor). Where it fits: shrinking before VAE encode, resizing an upscaled image back to a target resolution, or normalizing a batch of images to one size before they hit a model. Just keep expectations honest - this is a geometric resize, not an upscale. It makes pixels bigger or smaller; it doesn't invent detail. The KB's upscaling ladder has a whole ecosystem of real detail-adding upscalers, and this node is not one of them. Reach for it when you want dimensions, not when you want more resolution.

Install. Same pack install: ComfyUI Manager → search "isekai" → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/isekai-sh/isekai-comfy-node
cd isekai-comfy-node
pip install -r requirements.txt

Restart; it's under Isekai → Image/Transform. No models, pure Pillow.

Common issues. The Percentage default is the #1 footgun - change modes, forget to touch the sliders, and you get a 1% image. The batch gotcha applies (first image only, one image out - loop per-image for batches). And the house silent-failure pattern: an exception logs [Isekai] Scale Error to the console and returns the original image, so a resize that "did nothing" means it threw. One more honest note: this pack has no community footprint to lean on - it's a small utility pack, so treat the README and the source as your only docs, and when in doubt, preview before you rely on the dimensions.