-
-
Notifications
You must be signed in to change notification settings - Fork 4
Search Providers
| Source | What it searches |
|---|---|
| Local Database | Popular model entries and the local ComfyUI-Manager model-list.json database. |
| HuggingFace | Model repositories and files on Hugging Face. |
| CivitAI | Model pages, versions, file metadata, and hash matches. |
| CivArchive | Archived CivitAI model pages and file metadata. |
| LoRA Manager Archive | The local archive maintained by ComfyUI-Lora-Manager. |
The Everything search uses the sources enabled in Options. You can run a single provider search from the batch search menu. Each provider keeps its own result and status for every missing model.
Model Resolver searches by the workflow filename and normalized filename. It also handles model URNs and hash lookups for providers that expose exact file hashes. Local matching ignores case, file extensions, separators, and common naming changes.
Open Options > CivitAI to control the available methods:
- tRPC search is the main web search method.
- HTML fallback follows the public CivitAI page format.
- API search checks the public models API by name.
- CivitAI Models To Inspect limits how many candidate models Model Resolver checks for an exact file.
The CivitAI API key supports account access, model details, and downloads that require authentication. A CivitAI session token helps with pages that require accepted terms. Use the Check buttons after entering credentials.
Open Options > HuggingFace to control:
- Hugging Face API search.
- The ComfyOrg fallback index.
- Brave Search fallback for public Hugging Face links.
Use a read-only Hugging Face token for private or gated repositories. Brave Search requires its own API key. Model Resolver validates fallback links before it offers them as download sources.
Search results show provider status separately from an empty result:
-
unavailablemeans the provider or network path failed temporarily. -
rate_limitedmeans the provider asked the client to slow down. -
not_foundmeans the provider returned no matching page. -
errormeans the provider rejected the request or returned another failure.
When a provider marks the error as retryable, click its retry action. Model Resolver reruns that provider with a fresh request and leaves other search results unchanged.
Model Resolver saves server settings in model_resolver_settings.json inside the custom node directory. The file contains API keys and tokens, so keep it private and do not commit it to Git. The repository ignores this file.