Flip

IsekaiFlip mirrors an image horizontally, vertically, or both. The humblest node in the pack, and also one of the most honest: a one-dropdown tool that does one thing and does it correctly. You'll reach for it when text in a generated image comes out backwards, when a composition's light source is on the wrong side, or when you want to quadruple a dataset by mirroring every image.

How it works

It converts the image to PIL and applies Image.transpose with the right flip for your direction:

• Horizontal - mirrors left-to-right. The classic fix for backwards text and reversed signage. Also the standard data-augmentation trick, because a horizontally flipped image is still a plausible image in almost every scene.
• Vertical - mirrors top-to-bottom. Turns the world upside down; useful for reflection effects or when something's genuinely inverted.
• Both - flips both axes, which is the same as rotating 180°. Handy if you want a full inversion without rotating through the interpolation that a rotation node would apply.

No parameters beyond the dropdown, no sliders, no surprises. It's exactly as fast as you'd hope - a pure pixel remap, no resampling, nothing to tune.

The inputs

• image - the image to flip.
• direction - Horizontal, Vertical, or Both.

Output is a single image tensor.

Installing it

Part of isekai-comfy-node:

ComfyUI Manager: search "isekai" → install isekai-comfy-node → restart. Manually:

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

Restart and it's under Isekai → Image/Transform. Pillow and requests only; no models, no downloads.

Common issues

There is almost nothing that can go wrong here, which is the point. The one thing to remember is the pack-wide batch caveat: a batched image only gets its first frame flipped. For single images - which is basically always - it just works.

Worth thinking about once: if you're using a flip to fix backwards text, check whether the underlying model itself was trained with a mirrored bias before you blame the node. Some models consistently generate text the wrong way round, and the flip node is a perfectly good band-aid for that - just know you'll be applying it every time, so you might as well put it permanently in the workflow.