Legion: Importer
The receiving end of every Legion worker workflow
- output_1
- output_2
- output_3
- output_4
- output_5
- data_exchange_root
LegionImporter is the front door of a LegionPower worker workflow. It sits at the very start, receives whatever your main graph sent through a Legion: Master node, and presents it as regular ComfyUI outputs you can wire into the rest of the worker. It's also the anchor the Master uses to inject its data-exchange path into the workflow - which is why it must be in every worker workflow, or the Master refuses to run at all.
The mental model: your main ComfyUI serializes its inputs to a temp folder and writes a manifest; this node reads that manifest and hands you deserialized tensors and primitives. You don't do any setup for it - the Master patches its data_exchange_root value automatically right before submitting the workflow.
Inputs and outputs
data_exchange_root(required) - the path the Master injects. Leave it alone; it'sforceInput, and the pack's docs say explicitly not to connect it manually.output_1…output_5- deserialized data, in the same order your Master'sinput_1…input_5were fed. These wire into your worker's processing nodes.data_exchange_root(output) - a passthrough you connect to LegionExporter's required input. This is what keeps the whole exchange consistent, and the Exporter's security check relies on it.
Supported data
Only what the pack's serializers know: images (a single tensor or a batch) and primitives (int, float, str, bool). Anything else is skipped with a warning, and the corresponding output comes back None. That's the pack's current boundary - latents, models, and conditioning are on the roadmap but not shipped in v0.1.
Installing
ComfyUI Manager: search "LegionPower". Manual:
cd ComfyUI/custom_nodes
git clone https://github.com/Transhumai/ComfyUI-LegionPower.git
cd ComfyUI-LegionPower
pip install -r requirements.txt
Restart ComfyUI. Just requests; no models to fetch.
Troubleshooting
Workflow '...' must contain a LegionImporter node- thrown by the Master when the worker workflow is missing it. It's not optional, it's load-bearing.Input manifest not found-data_exchange_rootis pointing somewhere stale or wrong. Usually the workflow file got edited and the Importer's value was disconnected; re-run through the Master to re-patch it.- Outputs arrive
None- you fed an unsupported type on the Master side, or the name mapping didn't line up. Remember the Master sendsinput_1..Nand this node serves them asoutput_1..Nin order.
Not the flashiest node in the pack, but the most necessary one. No Importer, no worker.
Inputs (1)
| Name | Type | Default | Description |
|---|---|---|---|
| data_exchange_root | STRING | — |
Outputs (6)
| Name | Type | Description |
|---|---|---|
| output_1 | * | — |
| output_2 | * | — |
| output_3 | * | — |
| output_4 | * | — |
| output_5 | * | — |
| data_exchange_root | STRING | — |