Simpler dummy run (middlek)
Run a model and get noise back — the smoke test node
- model
- IMAGE
Simpler dummy run (middlek) exists to answer one question: does this model actually run? It takes a MODEL in, shoves a tensor of random noise through it, and spits the result out as an IMAGE. The image is pure static - it's a smoke test, not a generation.
How it works
The mechanism is exactly what it sounds like. It grabs the first parameter of the incoming model to figure out which device and dtype the model lives on, builds a 1x3x256x256 random tensor to match, runs it through model(...), and permutes the channels so ComfyUI can display it as an image. That's the entire node.
So what's it for? Verifying the wiring in this pack:
- Feed it the
MODELfrom Memory safe wrap (middlek), and you confirm the wrapped model loads, moves to the right device, and survives a forward pass - the whole point of the wrapper. - Feed it the
MODELfrom Simple dummy model (middlek), and you've smoke-tested the whole chain with zero checkpoints.
If you get an image out (noise, yes - but an image), the plumbing works. If you get an error, you've isolated it to the model path rather than some unrelated workflow bug.
The honest caveats
- The output is random noise by construction. It uses
torch.randn, so don't look for meaning in the preview - the only signal is "it didn't crash." - It hardcodes 256×256 and a 3-channel input. Models that expect a different input shape will fail. That's a limit of the test node, not your model.
- It's not a benchmarking tool. It's a "does it run," not a "how fast" - and not a substitute for real inference.
Install
Part of the same no-frills pack, no dependencies beyond ComfyUI itself:
cd {your_comfyui_path}/custom_nodes
git clone https://github.com/MiddleKD/ComfyUI-mem-safe-wrapper
Restart ComfyUI, and both dummy nodes appear under the safewrapper category. ComfyUI Manager users can just search "ComfyUI-mem-safe-wrapper".
This is debugging scaffolding the author left in the build, and it's honest about being that: one tiny job, done plainly. If you're testing the pack, it's exactly what you want. Otherwise you'll never touch it - and you'll know you're not missing anything.
Inputs (1)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | — |