MPN Switch
One boolean to flip between two complete setups (model + prompts)
- model1
- positive1
- negative1
- model2
- positive2
- negative2
- model
- positive
- negative
MPN Switch is the pack's real workhorse, and its name tells you exactly what it holds: the full Model-Positive-Negative triple, twice. It's an A/B switch over everything a KSampler needs. Flip one boolean and you swap the model, the prompt conditioning and the negative conditioning at once - no rewiring, no bypass groups.
This is the generalized version of the pack's one documented node, the LoRA Switch. The README frames it plainly: you have model1/condition1 as the original model and prompts, and model2/condition2 as the model with a LoRA applied plus its trigger words appended. Switch OFF, you run clean; switch ON, the LoRA and its prompts are in the queue. MPN Switch is that same idea opened up: either side can be anything.
How it works
A straight boolean select. active OFF → outputs set 1 (model1, positive1, negative1); active ON → outputs set 2. Nothing is merged, mixed or sampled here - the chosen triple passes through untouched, which is exactly what you want from a switch. Set it by toggling the widget, or wire it from a Primitive/control node if you want to flip it mid-queue.
The inputs that matter
- active (BOOLEAN, default off) - the only knob. Off is set 1, on is set 2.
- model1 / model2 (MODEL) - the two model lines. For the classic LoRA case,
model1comes straight from a loader andmodel2is the output of a LoRA Loader fed by that same model. - positive1 / positive2, negative1 / negative2 (CONDITIONING) - the prompt pairs. The README's own advice for the LoRA branch: add the trigger words via a Conditioning (Combine) or Conditioning (Concat) node before the switch.
Outputs: model, positive, negative - wire them into your KSampler. Where this shines is A/B testing: two checkpoints with two different prompt styles, or "with LoRA" vs "clean", compared in a couple of clicks instead of a graph edit.
Installing it
ComfyUI Manager → Install Custom Nodes → search TGu Utilities → install → restart. No Python dependencies, no model downloads; it's a pure graph node, safe and offline-friendly. Or:
cd ComfyUI/custom_nodes
git clone https://github.com/TGu-97/ComfyUI-TGu-utils
Restart and it appears in the TGu_util category.
Troubleshooting
- Nothing changed when you flipped it. Check that both triples are actually wired - an empty slot on the inactive side silently does nothing until you switch to it.
- You want "pick whichever is connected" instead of a manual toggle. Different philosophy. This is a plain boolean; the rgthree ecosystem's Context Switch picks the first non-null input, which some people prefer when they're swapping models around.
- It's a tiny, unmaintained pack. The README covers one node and the GitHub repo is currently unreachable, so treat this as freeze-dried: it works as-is, but don't expect updates. For a node this small, that's usually fine.
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| active | BOOLEAN | false | — |
| model1 | MODEL | — | |
| positive1 | CONDITIONING | — | |
| negative1 | CONDITIONING | — | |
| model2 | MODEL | — | |
| positive2 | CONDITIONING | — | |
| negative2 | CONDITIONING | — |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |
| positive | CONDITIONING | — |
| negative | CONDITIONING | — |