🔄 Flip Image

Exactly what it says: it flips an image, either horizontally (a left-right mirror) or vertically (upside down). No mystery, no model, no settings beyond the direction. You could do this in any image editor, but the reason it's a node is that you often need it inside a workflow - flip a reference before it hits a ControlNet, mirror a generation for a symmetry pass, or build a quick augmented set - and bouncing out to Photoshop for a mirror is silly.

It's a leaf utility. The interesting thing about it isn't the node, it's the handful of genuinely useful reasons to flip a picture mid-graph.

How it works

It reorders the pixels along one axis. Horizontal flip mirrors columns (the left edge becomes the right edge); vertical flip mirrors rows (top becomes bottom). It's lossless - no resampling, no quality change, the same pixels in a new arrangement - and instant. One IMAGE in, one flipped IMAGE out.

The inputs and outputs that matter

There are only two inputs and one output:

• image - what you're flipping.
• flip_direction - horizontal (the default, and the one you'll use 95% of the time) or vertical.

The output is flipped_image, ready to wire into whatever's next.

Why horizontal is the useful one: a left-right mirror keeps the image looking natural - most scenes read fine either way - which is why it's the standard move for data augmentation (double your training images for free) and for breaking a model's directional bias when a subject keeps facing the same way. Vertical flips look unnatural for most photographic content, so they're mostly for textures, reflections, and deliberate effects.

How to install it

ComfyUI Manager: search SDVN_Comfy_node, install, restart. Manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/StableDiffusionVN/SDVN_Comfy_node

then pip install -r custom_nodes/SDVN_Comfy_node/requirements.txt from your ComfyUI root, and restart. It's under SDVN → Image. There's nothing to download and no key to configure.

Common issues & troubleshooting

Text in the image reads backwards. That's a horizontal flip working correctly - mirroring reverses any text, logos, or numbers in the frame. If you need the picture flipped but the text readable, that's a job for an editor with layers, not a whole-image mirror.

Faces look "off" after flipping. Human faces are subtly asymmetric, so a mirrored portrait can look uncanny to people who know the subject. Nothing's broken; it's just what mirroring does. For augmentation this doesn't matter; for a hero shot of a specific person, it might.

I flipped it but the downstream result didn't change. If you're feeding a flipped image into a ControlNet or img2img and expecting a different composition, confirm the flip node is actually in the path (its output wired in), not sitting disconnected off to the side. It's an easy one to leave dangling.