KIE • Inspect API Definition (Advanced)

This node does nothing. It submits no job, costs no credits, and produces no media. It reads the model catalog that KIE.ai Nodes Next built from KIE's own API documentation and prints the boring-but-decisive parts of one entry: the HTTP method, the endpoint path, a link to the docs page, the example request body, and the default model string. That's it, and that's why it's under KIE Next/Advanced rather than the model folders.

Why would you want that? Because every other node in this pack is generated. On startup, the pack reads KIE's public API documentation index and mints a ComfyUI node for each callable model page, then enriches each one from that model's own docs. The knobs you see on a Seedance or Seedream node are the pack's translation of a request schema you never see. When a model behaves oddly - a field does nothing, a model name looks unfamiliar, a node shows up that nobody's tutorial mentions - this is the node that tells you what's underneath instead of guessing.

Inputs and outputs

One input, five outputs, all plain STRINGs.

operation is a dropdown of every API operation the pack discovered - the bundled catalog that ships with v0.5.0 carries 241 of them, from Chat Models > Claude • Claude Fable 5 down through the ByteDance, Kling, Wan, Runway, Suno and utility endpoints. Pick the one you care about.

Out comes method (POST, usually, sometimes GET), endpoint (the path on KIE's API, e.g. the Suno jobs route or the file-upload route), docs_url (the .md page the pack read to build the node), example_body_json (KIE's own example payload, pretty-printed), and default_model (the model string the pack folds into requests when the docs specify one).

Those are text outputs, so you can do two things with them. Read them in the UI - in most frontends, string outputs can be shown in a widget preview - or wire them somewhere. example_body_json is the useful one: it's exactly what the schema-less expert override fields want (the expert_override_json widget on endpoints whose docs publish no request schema yet), and it's the same JSON shape the pack's own universal/advanced task node takes when you need a KIE endpoint that has no dedicated node.

Installing it

Nothing here changes the install: Manager, search KIE.ai Nodes Next (publisher id felipeosilva); if Manager finds nothing - the README still describes registry publication as prepared rather than shipped - clone it:

cd ComfyUI/custom_nodes
git clone https://github.com/felipederosilva/ComfyUI-KIE-Nodes-Next
pip install -r ComfyUI-KIE-Nodes-Next/requirements.txt

Restart ComfyUI afterwards. There are no model downloads: the dependencies are requests, Pillow, numpy and PyYAML, and the actual model weights live on KIE's servers, which is the entire premise of the pack and also its entire catch (see the cost and privacy notes below). The one-time API key goes in ComfyUI → Settings → KIE.ai Nodes Next → Connection; KIE_API_KEY in the environment takes priority for headless boxes.

Gotchas

An empty or short list means your catalog is stale or thin. The dropdown is filled from the catalog file your install synced - the whole reason a node for the hot new model can appear without a plugin update, and the reason your node list may not match a screenshot someone posted last month. If an operation you expect isn't there, run KIE • Refresh Model Catalog first; the pack also re-checks for doc changes at most once every 24 hours.

Missing endpoints are normal. Some doc pages describe calls the pack couldn't resolve to a real method/path (method and endpoint come back empty). Those are coverage holes from KIE's docs, not from your install. Treat them as a signal to check the docs page link rather than as a bug to file.

It is not a free preview of a paid call. You can read a request definition all day; this tells you what would be sent. Nothing is sent. For that you need one of the actual model nodes, and those are metered - an API node is the right tool for a model you can't run locally, and the wrong default for one you can. Every model node reports credits spent and credits left after it runs, which is a lot more honest than most API plumbing in this ecosystem.