KIE • Refresh Model Catalog

Here's the problem this node exists to solve. KIE.ai isn't a model - it's a storefront that resells a few hundred hosted models, and it adds and retires them constantly. A pack that hard-codes its node list at release time is out of date within a month, which is exactly the failure mode the older generic Any KIE API node had: one hopeful JSON field and a prayer. So KIE.ai Nodes Next inverts it. On startup the pack reads KIE's public English API documentation index, builds a node for every callable model page it finds, then enriches each node from that model's own docs. Your node menu is a mirror of the catalog, not a snapshot of the author's release day.

KIE • Refresh Model Catalog is the manual trigger for that pipeline. Wire it up, queue the graph, and it re-reads the index. That's the whole job.

Inputs and outputs

One input: refresh (BOOLEAN, default true). Leave it on and the node performs a real sync; switch it off and it just reports the catalog summary it already has on disk without touching the network - handy for checking what you're running without waiting on a fetch.

Three outputs. status_json is the useful one: a pretty-printed report with the sync result plus catalog stats - when it was generated, how old it is, how many models and operations it holds, how many operations resolved to a real method/path, how many documentation pages were indexed, scanned and failed, and whether the deep sync completed. models and api_operations are the same two counts as plain integers, so you can eyeball "did the catalog actually grow" without parsing anything.

This node is registered as an output node, meaning it runs when you hit queue even with nothing wired downstream. That's deliberate - you can drop it in an empty workflow and just run it.

It never spends credits. It reads public documentation, writes a cache file, and that's the entire operation.

What the pipeline actually does

A shallow pass registers every documentation page as a catalog entry first, then the deep sync visits each page to pull its request example and parameter metadata. Doing it in that order matters: a page that fails to fetch keeps its entry instead of vanishing from your node menu. Model variants that KIE documents separately become separate nodes - that's why Seedance 2.0, 2.0 Fast, 2.0 Mini and 2.5 are four nodes with four different parameter sets, and why the model selector you'd expect doesn't exist.

The pack ships an offline bootstrap so the first launch isn't empty (Seedance 1.5 Pro, 2.0, 2.0 Fast, 2.0 Mini, 2.5, Claude Opus 5, GPT 5.6 Sol, Suno Generate Music, Veo 3.1 Generate Video, plus the utility endpoints). The first deep sync then expands that to the full catalog and reloads the ComfyUI frontend once so the new nodes actually appear in your menu.

Install

Manager first - search KIE.ai Nodes Next (publisher id felipeosilva). If it isn't there yet, the README's registry publishing is described as prepared, not shipped, so:

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

Restart, then set the API key once under Settings → KIE.ai Nodes Next → Connection. The sync itself needs no key; only the generation nodes do. Windows users have a one-click .vbs installer in dist/, and a matching full uninstaller - use the uninstaller rather than deleting the folder by hand if you ever want a clean slate, since it also cleans up node backups.

Gotchas worth knowing

You don't usually need to run this. The pack already checks for catalog changes at most once every 24 hours, and the startup pass handles new models. Reach for this node when a model you read about isn't in your menu, when a node's fields look wrong for the model, or right after a KIE announcement. It's also the fix for the general "why don't my nodes match this tutorial" problem, which is chronic in the API-node category because the model roster changes faster than any written guide does.

A sync in flight is a no-op, not an error. If two calls collide, the node reports busy and returns the summary. The 24-hour interval is per-install, not per-queue, so hammering it won't make new models appear faster.

Backups inside custom_nodes were a real bug once. v0.3.0's Windows installer left old backups inside the node directory, where ComfyUI would happily import them as active nodes and give you doubled-up entries. v0.3.1 quarantines them outside the node folder. If you're upgrading an old install and see duplicate or ghost nodes, suspect a stale backup directory before blaming the catalog.

And the honest caveat: this pack is essentially invisible on Reddit. Searching the corpus for "KIE-Nodes-Next" returns zero threads - there is no community troubleshooting thread to lean on, so the README and the shipped source are your only documentation. That's typical for the reseller-wrapper niche: the model roster changes so fast that nobody writes it up, and the vendor's own docs are what the pack reads anyway.