⬡ Polyhedron Select Model Switch
Six models in one dropdown, swapped without a reload
- model_name
Most ComfyUI users discover the model-swap problem the hard way: you want to A/B two checkpoints, so you either edit the loader's dropdown between runs or duplicate the entire loader-and-sampler half of the graph. The Polyhedron Select Model Switch is the answer to both - a central model selector with six slots, and any one of them can be chosen by a single integer.
The mechanism is the part that makes it good. It forwards the pick as a filename string, not as a loaded model. Nothing is read from disk until a loader downstream actually asks for it, so switching slots costs a dropdown change, not a reload - and no model is hogging VRAM while it's sitting in an unselected slot. Each of the six slots lists every known model folder in one dropdown - unet, diffusion_models, checkpoints, and the GGUF variants - so your Q8 Wan file and your fp8 UNET sit in the same list.
How it wires up
It uses the classic AnyType trick, so its single model_name output plugs into any loader's COMBO input. The pattern is:
ULSModelSwitch (model_name) → UNETLoader (unet_name) → MODEL → the rest of your graph
or a GGUF loader's unet_name, or a checkpoint loader's ckpt_name - whichever file-type slot you're switching. select is the widget that matters: 1–6, and it can be a conversion-to-input or driven by anything that produces an int.
One warning that the pack's own source raises explicitly: don't wire the switch's output into the LoRA Stack's model input directly. The switch hands out a filename string, and a filename where a MODEL is expected poisons every consumer downstream until an unrelated node crashes. The Stack raises a loud error telling you exactly that - feed the switch into a loader's combo input, and wire that loader's MODEL output into the Stack. The tooltip on model_1 through model_6 is just "choose from known model files", which undersells the one real quirk: an empty slot prints a warning and returns nothing, so check that the selected slot actually has a file before queueing.
Installing it
It's part of the Polyhedron Suite pack, so one install covers all of it:
cd ComfyUI/custom_nodes
git clone https://github.com/PolyhedronAI/ComfyUI-PolyhedronLoRAStack.git
# restart ComfyUI
Or via ComfyUI Manager, search "Polyhedron Suite". No extra dependencies, and since this is a plain-widget node it renders fine under the new "Modern Node Design" renderer that the hand-drawn Stack and Engine nodes don't.
The honest limitation: this is a file-name switch, so it can't swap between two different architectures or two different VAE configs mid-graph - it picks a filename from the folders ComfyUI already knows. If you need that, you're into conditional-graph territory. But for the 95% case - trying a few checkpoints against the same LoRA stack - it's the cleanest six-slot answer around.
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| select | INT | 11–6 | Which slot to use (1–6). |
| model_1opt | COMBO | Slot 1 — choose from known model files. | |
| model_2opt | COMBO | Slot 2 — choose from known model files. | |
| model_3opt | COMBO | Slot 3 — choose from known model files. | |
| model_4opt | COMBO | Slot 4 — choose from known model files. | |
| model_5opt | COMBO | Slot 5 — choose from known model files. | |
| model_6opt | COMBO | Slot 6 — choose from known model files. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| model_name | * | — |