Clippy Reloaded (Load Image from Clipboard)

You know the loop. You copy an image out of a browser or a screenshot tool, save it somewhere, drag it into a Load Image node, and then spend the next minute wondering where you actually saved it. Clippy Reloaded exists to delete that loop. ClippyRebornImageLoader grabs whatever image is on your clipboard the moment you queue, and feeds it into your workflow as a normal IMAGE - no save step, no file path, no nothing. Copy the picture, hit Queue Prompt, done. The name is a full-on joke, but the node is genuinely the fastest way to get a reference image out of a screenshot or browser tab and into a pipeline.

It ships in a tiny standalone pack by the author of the original "clippy" node that used to be buried in a bigger pack where nobody could find it. It's now on its own with a dedicated folder, one __init__.py, and the emotional support paperclip support group. The whole point of the pack is personality: every queue, Clippy posts a snarky line to your console and to the node's own UI, with over 80 different messages. Most are mildly insulting. That's a feature.

How it works

Under the hood it's doing exactly one real thing: PIL.ImageGrab.grabclipboard(). PIL ships with ComfyUI's Python environment, so there are zero extra dependencies - no requirements.txt, no model downloads, nothing to fetch. The node's IS_CHANGED returns nan, which is the ComfyUI trick that forces re-execution on every queue, because the clipboard may have changed since last time. It can't cache a result and it shouldn't.

The interesting part is the edge-case handling. Real clipboard images can come in as RGBA or palette mode, so it composites transparency onto a white background and converts everything to plain RGB - which is why the output is always described as "Image from clipboard (RGB format)." That means transparent PNGs pasted in get a white backdrop rather than a weird alpha artifact downstream. And if you grab a file in your clipboard manager instead of pixels, it still tries to open the first path as an image. If the clipboard is empty or holds text, it doesn't error out - it outputs a small grey placeholder and a message about Clippy being disappointed, so your workflow keeps running instead of dying mid-batch.

The output that matters

This is an output node with a single output worth wiring: image, typed IMAGE. That's the same type Load Image produces, so it plugs into anything - a VAE encode, a ControlNet preprocessor, an upscaler, img2img's latent path, whatever. There are no inputs at all. No settings. The README is blunt about this: "Clippy didn't think you needed any."

The node also shows a preview of the grabbed image right inside its own UI, which doubles as a sanity check that the clipboard actually had what you thought it had.

Installing it

Two ways, both trivial:

# ComfyUI Manager: search "Clippy Reloaded" and install
# or manually:
cd ComfyUI/custom_nodes
git clone https://github.com/shootthesound/comfyui-clippy-reloaded

Restart ComfyUI, then add Clippy Reloaded (Load Image from Clipboard) from the image category. No extra pip packages, no model files - it's pure standard-library plus the torch/PIL/numpy that ComfyUI already loads. In a world where the top complaint about this ecosystem is dependency hell, that's a genuinely nice property.

Common issues & gotchas

Most "problems" here are expectations, not bugs. If nothing appears, you queued with an empty clipboard - that's the placeholder case, not a failure. Paste the image, requeue.

• It reads the clipboard on queue, not on add. Copy the image before you queue, not while you're staring at the node. This trips everyone once.
• A clipboard that holds a file path (some clipboard managers, "Copy as path") is handled by opening the first file in the list - so it usually still works.
• Not an image in the clipboard (text, or a file ComfyUI can't decode): you get the grey placeholder and the workflow proceeds, so check the node's preview when output looks wrong.
• Remote/headless setups where the server has no real clipboard session can return nothing - this one expects a desktop with an actual OS clipboard behind the API.

It's tested on Windows; macOS and Linux use the same PIL call and should behave, though the author admits to not having a Linux box to verify. For a utility node this small, that's about as much as you need to worry about. Clippy will handle the rest, and tell you about it.