Image Mirror

ImageMirror flips an image. That's the whole node. You feed it an IMAGE tensor, pick horizontal or vertical, and you get the mirror back. There's no secret sauce, and honestly there doesn't need to be - in a pack that's mostly about image prep, it's the utility you grab when you're assembling a reference sheet and you want a face or a composition mirrored to keep direction consistent.

How it works

The node converts the tensor to numpy, flips along the right axis, and converts back. Horizontal flips along the width axis (left↔right, so text and asymmetrical faces reverse), vertical flips along the height axis (up↔down). It operates on the whole batch, so a video or multi-frame batch gets every frame flipped consistently, which is the one thing you'd actually care about here.

It's functionally identical to core ComfyUI's Image Flip node. If you're already running YCNodes for the IC combination stuff, using this one keeps your graph's node provenance uniform - that's a real reason to reach for it, just not a technical one. If you're not already on this pack, the core node does the same job and you don't need to install anything.

Inputs

• image - any IMAGE tensor.
• mirror_type - horizontal (default behaviour for most "flip" use cases) or vertical.

Output is a single IMAGE, same shape as the input, no resolution change. Wire it anywhere a flipped image makes sense - before compositing, before a tiled upscale (some textures mirror better than others), or into a reference grid where a mirrored copy helps the model read symmetry.

Installing it

It ships in ComfyUI-YCNodes. Via ComfyUI Manager, search ComfyUI-YCNodes and install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/yichengup/ComfyUI-YCNodes

Restart ComfyUI afterwards. The pack pulls in torch, numpy, pillow, opencv-python, and scipy; this particular node only actually uses torch and numpy, so it's not a dependency risk.

Gotchas

Only one, and it's conceptual rather than technical: flipping is not the same as rotating. A horizontal flip mirrors the image; rotating 180° turns it upside-down. If your mirrored text looks wrong, that's physics, not a bug. And keep in mind the node mirrors the entire frame - if you only want part of an image mirrored, crop or mask first, flip, then composite, because there's no region parameter here.