🛑 DUMMY ECHO Loader

The name says "CheckpointLoaderSimple," so you'd expect it to drag a model into your workflow. It doesn't. This node takes one input, returns three outputs, and every single one of them is literally None. It's a fake - and that's the whole point.

What this actually is

This is the DUMMY ECHO Loader from the ComfyUI-Dummy_Node_Pack, a small pack of placeholder nodes the author calls "Nodos Fake." Its job isn't to load checkpoints; it's to make a broken workflow load. ComfyUI has a real habit of refusing to open a JSON when a node class is missing - you get the red missing-node banner, the graph won't run, and sometimes the canvas looks like a car wreck. That's the single most common complaint in the ComfyUI ecosystem: you download a workflow, and half its nodes are packs you don't have.

This pack exists for the other half of that problem: the workflow references a node that no pack you'll find provides, often because it was exported from Automatic1111 or a cloud tool. ECHOCheckpointLoaderSimple is one of those. The pack registers the class under that exact internal name, so when ComfyUI reads the JSON it thinks the node exists. The graph loads, the cables stay connected, and you can finally see where things went.

How it works

Grep the source and there's no mystery:

class Fake_ECHOCheckpointLoaderSimple:
    RETURN_TYPES = ("MODEL", "CLIP", "VAE")
    FUNCTION = "do_nothing"
    def do_nothing(self, **kwargs):
        return (None, None, None)

That's the whole node. It declares a MODEL, CLIP, and VAE output so the wire types match what the workflow expects, then hands back nothing. It doesn't touch your models folder, it doesn't read ckpt_name, and it imports torch without ever using it. Zero dependencies beyond ComfyUI itself, no requirements.txt, no model downloads - this pack is as light as a placeholder can be.

The trap to keep in mind: a None output is not a model. If you wire this into a real KSampler and hit Queue, it fails at execution time. The dummy is for reading the graph, not running it.

Inputs and outputs that matter

There's exactly one input:

• ckpt_name - a plain text field defaulting to fake_model.safetensors. Don't go hunting for that file; it doesn't exist. The widget is kept only so the workflow JSON doesn't lose the value when it reloads.

Outputs are the classic loader trio - MODEL, CLIP, VAE - which is your biggest hint about the fix. The native Load Checkpoint node has the exact same three outputs. That's your replacement: swap the dummy for Load Checkpoint, pick your real model, and reconnect.

Installing it

The README's method is a plain clone:

cd ComfyUI/custom_nodes
git clone https://github.com/FuryNocturn/ComfyUI-Dummy_Node_Pack.git

Then restart ComfyUI. It may also be findable through ComfyUI Manager if you search for "Dummy Node Pack," but the clone is the documented path and it's fast - no pip install step, no models to grab. One note: the README's own clone example has a placeholder username in it, so use the URL above, not the one in the readme.

Common issues

The big one is confusion about what "works" means. If the dummy loads the workflow but execution errors on it, that's not a bug - it's the node doing exactly what it's designed to do. The fix order is: replace with Load Checkpoint, reconnect MODEL/CLIP/VAE, then delete the dummy. You'll spot them easily because every node in this pack wears a 🛑 DUMMY prefix.

If you still see the missing-node error, the pack isn't actually installed - check that it cloned into custom_nodes and that you restarted ComfyUI after.

This isn't a node you'd reach for on a normal day. It's a rescue tool for that one frustrating Saturday where a promising workflow won't open. Install it, let it hold the wires in place, swap in the real nodes, and move on with your life.