upscale image to side 放大图片

This is the pack's older image-scaler, and you should know two things about it before you use it: it does the same side-based resizing job as Image_Scale_To_Side (2lab), and its side_length default is 1. Read that again. If you drop this node into a workflow, wire in an image, and hit run without touching anything, you're asking it to scale a side of your image down to a single pixel. You'll get a technically-correct, completely useless 1-pixel-tall render and a confused afternoon.

That default is the sort of thing that makes you suspect the author left a test value in and never looked back. It's the single most important thing to know about this node, because nothing else about it is surprising.

The inputs that matter:

• image - what you're scaling.
• side_length - the target length. Default 1, range technically the full 64-bit integer span. Set this. Always set this. Nobody wants a 1-pixel image.
• side - Longest, Width, or Height (note: no Shortest, unlike the newer node).
• upscale_method - a reduced set: nearest-exact, bilinear, area. No bicubic, no lanczos, no bislerp. For upscaling you're left with the two soft options and the pixel-y one; for downscaling area is still your friend.
• crop - disabled or center.

Output is a single image. It's flagged as an output node, which is a slightly odd choice for a scaler but harmless - it means ComfyUI will run it as a terminal even with nothing downstream.

So which one should you use? Image_Scale_To_Side (2lab). It's the same pack, the same side-based concept, but it has the full set of resampling methods, adds Shortest as a side option, and defaults side_length to a sane 1024 instead of a trap. This node appears to be the earlier draft that was never retired when the better one shipped - a recurring theme in this pack. If a downloaded workflow forces you onto ImageScale_Side, you can use it, but treat the default with the suspicion it deserves.

Installing it

Part of AI2lab/comfyUI-tool-2lab. ComfyUI Manager → search "comfyUI-tool-2lab" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/AI2lab/comfyUI-tool-2lab

Restart ComfyUI. No model downloads, no network - pure local tensor work.

Common issues

The 1-pixel default is the headline. Second: the reduced upscale_method list means you can't get bicubic or lanczos here even though they exist elsewhere in the pack - if you need them, use the other scale node or core ImageScale. And because Shortest isn't an option, "fit the smaller dimension to a value" isn't expressible here. Everything else follows the same rules as the newer node: center crop trims edges, area is right for downscaling, and don't expect detail creation from any of these - they resize, they don't upscale.