AIHub Meta Export Lora

If the expose nodes are the front door of the otavanopisto ComfyUI-aihub-workflow-exposer pack, AIHubMetaExportLora is the publishing desk. It's an output node that takes one of your LoRA files and writes a small JSON catalog entry plus an optional thumbnail into ComfyUI/aihub/loras - which is exactly what a connected client app reads to know which LoRAs exist, what they're for, and which models they fit.

The mental model: this pack doesn't let a client scroll your entire models/loras folder. It serves a catalog, and this node is how a LoRA gets into that catalog. Run it once and you've registered the LoRA; from then on, clients can offer it in their UI.

The inputs that matter

• lora - the LoRA file, picked from a dropdown populated by your models/loras folder. The node derives the catalog id from the filename (no extension). Leave it empty and it errors with "No lora specified."
• name / description - the human-readable bits. name falls back to the filename if empty; description is what shows in the client UI.
• context - image, video, audio, 3d, or text (default image). Tells clients which workflows this LoRA applies to.
• limit_to_family - the model family (default sdxl), like flux, qwen, sdxl. limit_to_group - a version-less model tag. limit_to_model - a specific model file. These filters are how the client decides whether this LoRA can be applied to the selected model, so getting them right is the difference between "the LoRA shows up everywhere" and "the LoRA only appears where it works."
• default_strength - the strength the client applies first (default 1, 0–10).
• use_loader_model_only - if true, the LoRA patches only the model, not the CLIP (the README's LoraLoaderModelOnly path). Some LoRAs, particularly certain styles, break conditioning if applied to the clip too.
• optional_image - an IMAGE input for the thumbnail. Provide one and the node saves {id}.png beside the JSON.

There are no outputs - it's a fire-and-forget output node.

What it actually writes

A {id}.json with those fields, a {id}.png if you fed it an image, and a default locale entry under aihub/locale/loras/default/ (that's the localization mechanism the pack uses for clients that send a locale header). Deleting or editing entries is done by hand on the filesystem - the README is explicit that these nodes only add, to prevent mistakes; they're not a full management UI.

Gotchas

The filter fields are the trap. Set limit_to_group without a limit_to_family and the node raises an error (family is mandatory if group is set). Set limit_to_family wrong and your LoRA silently won't appear for the models you actually use. And note the author's own advice: exporting a LoRA is only necessary when a workflow should accept any LoRA in a family. If your workflow is hardcoded to one model, you can load LoRAs directly with AIHubUtilsLoadLora and skip the catalog entirely.

Install - the pack has no requirements:

cd ComfyUI/custom_nodes
git clone https://github.com/otavanopisto/ComfyUI-aihub-workflow-exposer

Restart, and the aihub/loras folder is created for you (or make it manually per the README). This is the most "management-y" node in the pack, and for a workflow author it's genuinely handy - one run and a LoRA is client-selectable with a thumbnail and a description.