-
-
Notifications
You must be signed in to change notification settings - Fork 138
FAQ (Frequently Asked Questions)
Q: I'm getting strange errors, previews aren't showing, or model data looks wrong. What should I try first?
A: Try Rebuild Cache — it's the first thing to check when things go wrong.
LoRA Manager uses a caching system to pre-process model metadata, which dramatically improves startup speed and overall performance. Instead of scanning and parsing every file on each load, the cache serves previously collected data.
The cache can become stale in a few common scenarios:
- You manually edited
.metadata.jsonfiles alongside your models. - You changed the model directory structure, moved files, or added/removed models externally.
- The cache has been running for an extended period without a refresh.
When this happens, it can produce confusing symptoms — odd error messages, missing previews, incorrect thumbnails, or models not appearing where they should. Before asking for help or diving into logs, try rebuilding the cache first.
To rebuild the cache:
On the LoRAs, Checkpoints, Embeddings, or Recipes page, locate the Refresh button. Click the caret icon (▾) next to it — this opens a dropdown with additional actions. The second option, Rebuild Cache, will clear and fully regenerate the cache for that category from scratch.
Rebuild the cache on each affected page individually. No server restart is needed.
A: Some features, fixes, or improvements may only be available in the nightly version. If you'd like to try them before the next release, you can switch from the stable release to the nightly channel.
The steps depend on which version of ComfyUI Manager you're using:
If you're using the legacy custom node version:
- Open ComfyUI Manager.
- Search for LoRA Manager.
- Click Switch Ver. and select nightly from the options.
If you're using the new built-in Manager:
- In ComfyUI, click Extensions to open Nodes Manager.
- Search for LoRA Manager.
- Click the version number under the Version column.
- Select Nightly from the dropdown.
Once switched, the nightly version will be pulled and installed. You can switch back to the stable release at any time using the same steps.
Q: Why am I getting the error message "Invalid or missing CivitAI API key, or early access restriction" when trying to download models?
A: This error usually occurs for one of the following reasons:
- You haven't set your CivitAI API key correctly.
- You're trying to access a model that is currently restricted to early access users, and your account doesn't have the required permissions.
Make sure your API key is entered correctly in the settings, and check the model's access status on CivitAI.
A: There's no dedicated "unsorted" filter (and it's not planned), but you can find them using the existing folder tree on the left sidebar.
-
If you use Priority Tags with a path template like
/{model_type}/{first_tag}— open theno tagsfolder. That's where models land when nothing in your priority list matches their tags. - If you organize manually — click your root LoRAs directory in the folder tree. Any model file sitting directly there (not inside a subfolder) hasn't been moved into a folder yet.
A: No, you can't. The widget behaves very differently from a standard input.
The input widget in the LoRA Loader node is designed for quick referencing of available LoRAs and displaying a visual interface for users to easily view and control them. A key feature is that changes made in the widget are instantly detected by the node—without needing to run the workflow.
In contrast, a standard input only provides its value when the workflow is actually executed.
A: It depends on whether you've set a custom Recipes Storage Path.
If you haven't set a custom path — recipes are stored in the recipes/ folder under the first LoRA root directory (alphabetically). For example:
ComfyUI/models/loras/recipes/
If you've set a custom path (Settings → Recipes Storage Path) — recipes are stored at whatever path you specified.
Migrated ComfyUI and can't find your recipes? Look for the recipes/ folder under the LoRA root in your old ComfyUI installation, then copy the entire folder to the corresponding LoRA root in your new setup. If you were using a custom path, make sure that path still exists and is correctly set in your new environment.
Q: I run multiple ComfyUI instances with different model directories. Do they share LoRA Manager settings?
A: Yes — when Portable Mode is off (the default), all instances share the same cache and settings (~/.config/ComfyUI-LoRA-Manager/ on Linux). This means one instance may show models from another instance's directories until you manually refresh the model list.
To isolate settings per instance, run each instance with this environment variable:
LORA_MANAGER_PORTABLE=1 python main.py --port 8188When set, LoRA Manager stores settings.json, cache, wildcards, and backups inside the plugin's own repository directory instead of the shared user configuration folder. It also enables Portable Mode in the configuration file, so the setting persists — you only need the environment variable once.
A: When using the legacy (A1111) LoRA syntax format, LoRA tags reference models by file name only, so file names must be unique across all subfolders of your LoRA directory.
You can either switch to full path syntax, or use the Doctor's conflict resolver to rename the duplicates. See LoRA Syntax and Duplicate Filenames for details.
A: Yes, you can turn it off in either of these ways:
-
Use a command in the node - type
/noautocompletein the prompt field. Just typing/nois enough for it to appear in the suggestions; select it to turn autocomplete off. - Disable it in settings - go to ComfyUI Settings → Lora Manager, find Enable Tag Autocomplete in Prompt Nodes, and turn it off.
Both methods control the same setting, so either one works. Slash commands like /character, /artist, or /emb still work even when tag autocomplete is disabled.