Simple dummy model (middlek)

Simple dummy model (middlek) is not a model you'll ever generate with. It's a test fixture - a tiny torch.nn.Module with a single convolutional layer that the pack's author threw in so you can check whether the memory-safe wrapper actually works without loading a multi-gigabyte checkpoint.

What it does

The node has no inputs. Drag it onto the canvas and it hands you a MODEL output - a bare nn.Module whose forward pass is just one Conv2d layer over a 3-channel input. That's it.

It's genuinely useful in exactly two situations:

• Testing the pack itself. Wire it into Memory safe wrap (middlek) and then into Simpler dummy run (middlek). If that chain completes without OOM or attribute errors, your install is good - no SDXL download required.
• Prototyping a workflow's structure. When you're building a graph and just want the plumbing to exist while you figure out the rest, a dummy model lets you iterate on connections before you commit to loading a real checkpoint.

What it isn't

Don't mistake the output for something sampler-ready. If you feed it to a KSampler you'll get garbage or an error, because this model doesn't implement the diffusion-adjacent API a sampler expects. It produces a tensor and calls it a model - the point is the shape, not the sense.

Install

Comes free with the pack, and the pack has no dependencies beyond ComfyUI itself:

cd {your_comfyui_path}/custom_nodes
git clone https://github.com/MiddleKD/ComfyUI-mem-safe-wrapper

Restart ComfyUI, then look under the safewrapper category in the node menu. (Or install "ComfyUI-mem-safe-wrapper" via ComfyUI Manager.)

The one thing to keep in mind: this is scaffolding the author shipped for their own debugging, not a polished feature. It does exactly one dumb thing, and if you don't need a throwaway model, you don't need this node. That's fine - it costs you nothing to ignore.