ImageTransformTranspose

Sometimes you don't want a fancy rotate, you want a flip. ImageTransformTranspose is Allor's one-stop for every 90°-increment reorientation, and unlike the general rotate node it does the whole job with zero resampling. The pixels just move. No interpolation, no softening, no color bleed - which makes it the node to reach for when you're fixing orientation rather than composing an effect.

The seven methods

One dropdown, seven choices:

• flip_horizontally - mirror left/right
• flip_vertically - mirror top/bottom
• rotate_90 / rotate_180 / rotate_270 - clockwise quarter turns, all lossless
• transpose - mirror across the top-left↔bottom-right diagonal
• transverse - mirror across the other diagonal

If that last pair looks like a rotated flip, it is - a transpose equals a 270° rotation plus a flip, and it's the geometry behind "the image came out sideways, just fix it."

How it works

Mechanically it's the same Allor pattern: each frame tensor becomes a PIL image, gets Image.transpose applied, and comes back as a tensor, with the batch re-stacked. Because those operations (FLIP_LEFT_RIGHT, ROTATE_90, TRANSPOSE and friends) are pure remaps of existing pixels, there's no interpolation step and no quality cost whatsoever. You can flip an image back and forth all day and the pixels are bit-for-bit the same at the end.

That losslessness is exactly why you'd pick this over ImageTransformRotate for 90° turns. Rotating a photo that arrived sideways, flipping a mirrored composition before a detailer pass, or standardizing orientation across a batch for a training set - all jobs where you want the geometry fixed and the pixels untouched.

The inputs that matter

• images - the IMAGE batch (required)
• method - the seven-choice dropdown above

That's the whole interface. One input, one dropdown, one IMAGE output, same batch size as what went in. It wires into anything downstream that expects an image, since the tensor shape and range are unchanged.

Installing Allor

ImageTransformTranspose lives in the Allor Plugin, Nourepide's ~90-node image-processing pack. ComfyUI Manager is the easy route - search "Allor" - or clone it:

cd ComfyUI/custom_nodes
git clone https://github.com/Nourepide/ComfyUI-Allor

Restart ComfyUI. First run generates a config.json in the pack folder, and the bundled install.sh / install.bat handle requirements into your venv or embedded Python. The dependency list (rembg, onnx, plus onnxruntime via rembg) exists for the pack's segmentation nodes; this transform doesn't use them.

Common issues

The pack's known failure mode is updates. Its history was rebased to strip image files, the README warns this can break auto-updates, and users have reported Allor nodes silently disappearing after an update. A fresh re-clone is the reliable fix; "update_frequency": "never" in config.json keeps you out of that loop entirely.