Derp Clip Loader

The Derp Clip Loader is what a model dropdown would look like if it had grown up with self-respect. Instead of one clip_name combo box buried in a loader node, this thing hands you a deck - a panel of CLIP model cards you click to activate, drag to reorder, and hot-swap mid-workflow without ever touching a dropdown menu. It's part of the xcp_derp-UI pack (search "xcp_derp-UI" by lordwedggie in ComfyUI Manager), and it sits in the derpNodes menu alongside the rest of the derp loaders.

Why you'd bother

If you live on one CLIP model, skip this. But the moment you're A/B-ing SDXL against Flux, or you've got three text encoders you rotate for different styles, the value shows up: load all of them onto the deck once, then switching is a single click instead of a right-click-and-scroll ritual. The pack's docs are upfront about one thing the dropdown never tells you - the loader supports 23 different CLIP architectures, from stable_diffusion to ideogram4 and the newer krea2, "because the AI industry can't agree on anything." That's the whole reason CLIP loading needs its own node in the first place.

How it works

Here's the part that trips everyone up on first contact, so read it twice. The Derp Clip Loader is a pure virtual shell - check the node class and do_nothing literally returns None from Python. The real work happens in the JavaScript layer: when you pick a CLIP from the deck, the frontend packages a descriptor into the pack's wireless signal registry, and at execution time the xcpDerpSignalOut reconstruction engine resolves it into an actual loaded CLIP object. That's why the node has no input sockets at all and a single clip output of type CLIP. There's nothing to wire in - it broadcasts, and something else has to catch.

Which is where the pack's one hard rule comes in: every derp loader needs a derpRouter node somewhere in your workflow. The Router is the pack's central nervous system - it detects every wireless signal on the canvas, lets you add them as outputs, and routes them to receivers. Without it the Clip Loader is, in the author's own words, "just a very enthusiastic paperweight." Add the Router first, always.

The deck itself is straightforward: the ⚙️ settings toggle reveals the file browser so you can drop CLIP files onto the deck, plus the CLIP Type (architecture) and CLIP Device (default/GPU or cpu for when VRAM is tight) inline options. Toggle ⚙️ off once you're loaded and it hides the browser for a cleaner look. There's a refresh button so new models appear without restarting ComfyUI, clear/remove to wipe the deck or evict a single entry, drag-and-drop to reorder, and a search tab to filter.

Install

ComfyUI Manager → Install Custom Nodes → search xcp_derp-UI → install, restart. Or the manual way:

cd ComfyUI/custom_nodes
git clone https://github.com/lordwedggie/xcp_derp-UI-Release

Then restart ComfyUI. The good news: no requirements.txt, no pip installs, no model downloads. The pack ships its JS dependencies vendored, and it declares no Python dependencies beyond what ComfyUI already has. The only files this loader needs are the CLIP models you already own.

Where people get burned

The biggest confusion is expecting a wired output. Because the shell returns None and the CLIP arrives via wireless signal, the node only "works" when a Router is in the graph and the signal is routed to a receiver (another derp node, or a default ComfyUI node with a derp Wireless Extender attached). If you click a deck entry and nothing changes downstream, check the Router before you blame the loader.

One more honest note: this is a young, niche pack - at the time of writing it has essentially zero community documentation out in the wild, so the pack's own docs are your best reference. The flip side is the author actually maintains them, and the pack explicitly targets both the classic canvas and the Nodes 2.0 frontend that broke so many other packs. For that alone, if you're on the new frontend, it's worth a try.