🛑 DUMMY KSampler A1111

Look at the inputs on this thing and it reads like the real deal: model, seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, denoise. Ten inputs, one LATENT output, and the output is None. It never runs the sampler - there's no sampling happening anywhere in this node. That's deliberate.

What this actually is

KSampler_A1111 is a DUMMY placeholder from the ComfyUI-Dummy_Node_Pack. Some workflows floating around - usually ones exported from Automatic1111's webui or older cloud tools - saved a sampler node under the class name KSampler_A1111, which vanilla ComfyUI doesn't know. Normally that means a missing-node error and a workflow you can't even look at. ComfyUI's community has been complaining about exactly this "download a workflow, install missing nodes for an hour" loop since the start.

The pack's fix is simple and clever: register a fake class under that exact name. Now ComfyUI loads the JSON happily, the wires stay connected, and you can see the whole graph instead of a broken stump. The author's own README frames it as the chicken-and-egg problem - you can't fix a workflow you can't open, so this gets you to the fixing part.

How it works

The source is about as minimal as a node gets:

class Fake_KSampler_A1111:
    FUNCTION = "do_nothing"
    def do_nothing(self, **kwargs):
        return (None,)

The INPUT_TYPES mirror the native KSampler's signature so the graph's cables match up, then do_nothing returns a bare None for the LATENT. No model is touched, no diffusion happens, no seed is consumed. It's a scaffolding node - it holds the shape of your workflow in place while you rebuild it.

And here's the gotcha that trips people up: that None output is not a real latent. If you connect it into a VAEDecode and hit Queue, execution errors out. These dummies are for reading and repairing the graph, not running it.

Inputs and outputs that matter

The inputs mirror the native KSampler, so you mostly leave them alone. The ones worth knowing:

• sampler_name - defaults to euler; a plain text field here, not the dropdown you get on the real node.
• scheduler - defaults to normal.
• denoise - a float from 0 to 1, default 1, which you'd only care about for img2img once this becomes a real sampler.
• seed / steps / cfg - stored as widgets so the values survive the swap. cfg defaults to 8, which is the old-school Automatic1111 habit; if the model you actually want prefers lower, set it after you replace the node.

Output is a single LATENT, and it's the same shape as the real KSampler's - which is the tell for the fix. The native KSampler node has the identical input names, so you swap it in, copy over seed and steps, reconnect, and delete the dummy.

Installing it

Same pack as the DUMMY ECHO Loader, same one-line install:

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

Restart ComfyUI and you're done. No requirements.txt, no pip deps, no models to download - it's about as light as a custom node gets. ComfyUI Manager may find it if you search the pack title, but the clone is the documented route.

Common issues

The usual failure is thinking the dummy should generate something. It can't, and it won't - if you run it, the None latent breaks downstream. Replace it with the real KSampler first, then run.

Second: because sampler_name and scheduler are free-text strings here, a workflow could load with a misspelled sampler that the real node would reject. It doesn't matter while it's a dummy, but copy the values carefully when you swap, or the real KSampler will silently ignore an unknown name.

One more thing worth knowing: this pack is a repair tool, not a daily driver. If you're starting a fresh workflow, you don't need it. But the day a downloaded graph won't open because of a phantom KSampler_A1111, this is the five-minute fix that gets you back to actually generating.