π Model List (8 -> list)
The Model List That Loops the Graph
- model1
- model2
- model3
- model4
- model5
- model6
- model7
- model8
- MODEL
The switch nodes in this pack pick one of eight and pass it through. Model List (8 -> list) does the opposite and arguably the more useful thing: it gathers up to eight models into a list, and because ComfyUI expands list outputs into one execution per item, everything downstream runs once per model. Wire eight checkpoints in, hit queue once, get eight generations. That's the whole pitch, and it works exactly like that.
The trick is a single flag in the node's definition: OUTPUT_IS_LIST = (True,). Declaring that tells ComfyUI "my output is a list," and the engine responds by running the rest of the graph once for each element. The node itself does almost nothing - the make_list function collects model1β¦model8 into a Python list and drops any input that isn't connected. So you can wire four models and the list is just four items long; empty slots are skipped, not padded with None.
When you'd reach for it
Same prompt, same seed, same sampler - N different checkpoints. That's the clean use. You keep a single KSampler, one conditioning branch, and a list of models; one press produces one image per model, so you can compare how each checkpoint renders the same idea. The author's comment suggests wiring the list siblings by matching index - model1 paired with text1 from a Text List, clip1 with clip1 - so the index alignment survives the trip through the graph.
Inputs and output
model1β¦model8- optionalMODELinputs. Loaders only; leave unused slots empty.MODELoutput (list) - feeds themodelsocket of a KSampler or LoRA loader. WithOUTPUT_IS_LIST, ComfyUI runs that sampler once per model.
Category is Kyusu/list, separate from the Kyusu/logic switches so the two sets group apart in the node menu.
Install
Nothing to download, pure Python:
cd ComfyUI/custom_nodes
git clone https://github.com/TaIIo4KuH/kyusu-nodes
or ComfyUI Manager β "Kyusu Nodes", then restart.
The two gotchas that matter
An all-empty list runs nothing. If you wire zero models, the list is empty, and "once per item" means zero times - no image, no error, just silence. The first time that happens it's confusing. Check that at least one slot is connected.
All connected models load into VRAM. ComfyUI has to have every model in the list resident before it can run the graph once per item, so eight wired checkpoints means eight loaded models. For big checkpoints that's the practical ceiling on how many you can actually compare - a couple of Flux-scale models and the list gets expensive. This is a workflow-level convenience, not a memory optimization.
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| model1opt | MODEL | β | |
| model2opt | MODEL | β | |
| model3opt | MODEL | β | |
| model4opt | MODEL | β | |
| model5opt | MODEL | β | |
| model6opt | MODEL | β | |
| model7opt | MODEL | β | |
| model8opt | MODEL | β |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | β |