Image Crop Upload (Purz)

The frustrating thing about most "visual crop" nodes is that your crop disappears when you close the browser. PurzImageCropUpload is the version that doesn't: the image is uploaded to ComfyUI's input/ directory the moment you pick it, and the workflow serializes the filename plus every crop parameter - so reload the page, and your crop is still sitting there exactly where you left it. It's the persistent sibling of the client-side Image Crop node, and for anyone doing serious workflow iteration it's the one you actually want.

From ComfyUI-Purz's Purz/Interactive family, by PurzBeats (a Comfy Org staffer).

How it works

Unlike the client-side variant, the file goes to the server immediately and is processed there. The upload node exposes the crop controls as real, workflow-serializable inputs - image (the filename, picked from a dropdown of everything in input/), width, height, and the geometry settings - so saving your workflow saves your crop. Resizing happens server-side with Pillow (Image.LANCZOS), which is fast and deterministic.

The fit modes are the same four as the client-side version:

• stretch - squish the source to the output size, distorting as needed.
• cover - scale to fill the output, cropping the overflow, positioned by anchor.
• contain - scale to fit inside the output, letterboxing with background_color (default #000000).
• crop - like cover, but adds zoom and offset_x/offset_y on top so you can punch in further.

The inputs that matter

• image - the filename from your input/ directory. This is the one input everyone sets; everything else is optional tweaking.
• width / height - output size, 16 to 16384.
• fit_mode - stretch, cover, contain, or crop. Default is cover, which is the right default for "fill the frame."
• anchor - center, top, bottom, left, right, plus the four corners. Where the crop window sits within the source.
• zoom - 0.1–10, only meaningful in crop mode. This is your "punch in" knob.
• offset_x / offset_y - −1 to 1, fine positional nudges (in crop mode). These are what the draggable overlay of the client-side node updates for you.

Output is image, type IMAGE, at exactly the size you asked for.

Installing it

ComfyUI Manager - search ComfyUI-Purz - or:

cd ComfyUI/custom_nodes/
git clone https://github.com/purzbeats/ComfyUI-Purz.git
cd ComfyUI-Purz
pip install -r requirements.txt

Restart ComfyUI. Requirements are OpenCV, Pillow, torch, numpy - all standard, no model downloads.

Where people get burned

Remember which mode you're in. zoom and offset_x/offset_y do nothing in stretch, cover, or contain - they only apply in crop mode, so people drag those sliders, see no change, and assume the node is broken. Also, since the file now lives in input/, it counts against your ComfyUI storage and gets picked up by any node that scans that directory - nothing to clean up, but worth knowing where your disk space went. And if you're on the client-side Image Crop for a quick one-off, the same UX is there - this node just adds persistence and a slightly less drag-friendly interface.