CheckpointsBase diffusion models used as the primary generation backbone.
CheckpointLoaderSimple, UNETLoaderLoading...
Loading...
Documentation
Browse every model and LoRA you have ever used in one place. Numonic auto-detects models from your ComfyUI workflows and builds a searchable library with linked assets and version history.
The Model Library is populated automatically when you import ComfyUI assets. There is no manual setup required. Once your assets are imported, navigate to the library to explore your models.
/app/modelsNavigate to the Model Library to see every model detected across your imported assets. Use the type filter and search bar to narrow the list.
/app/models/[hash]Click any model in the list to open its detail page. Three tabs are available: Overview, Versions, and Assets.
Model detection happens during asset import. There is no separate step to run or configuration to enable. The process works the same way regardless of how you import your assets.
Upload or sync ComfyUI images via Smart Import, Connected Folders, or Desktop Sync.
Numonic parses workflow JSON and identifies every model, LoRA, and embedding referenced in each node.
Models are promoted to the library with linked assets and deduplication by hash. No manual entry needed.
The following model types are detected from ComfyUI workflow metadata. Each type is identified by the loader node that references it.
CheckpointsBase diffusion models used as the primary generation backbone.
CheckpointLoaderSimple, UNETLoaderLoRAsFine-tuned adaptors applied on top of a checkpoint for specific styles, characters, or concepts.
LoraLoader, LoraLoaderModelOnlyEmbeddings / Textual InversionsLearned token embeddings that modify prompt interpretation without changing model weights.
Referenced in prompt text as embedding:nameControlNet ModelsConditioning models that guide generation using structural inputs such as edges, depth maps, or poses.
ControlNetLoader, ControlNetApplyVAE ModelsVariational autoencoders that handle encoding and decoding of the latent space.
VAELoaderOther Node-Referenced ModelsAny model file referenced by a ComfyUI node that is not covered by the categories above.
Various loader nodesEach model has a dedicated detail page at /app/models/[hash]. The page is organised into three tabs.
Displays the model name, type, description, and tags. All metadata fields are editable: click any field to edit it inline. Changes are saved immediately and stored as Data Vault satellites, so the full edit history is preserved.
Shows version history for the model, listing each distinct version that has been referenced across your imported assets. Useful for tracking when you switched between model versions or identifying which version produced your best outputs.
A gallery grid of every asset generated with this model. Uses the standard asset grid with signed thumbnails. Click any thumbnail to open the full asset detail page.
The Model Library list view provides two ways to narrow down your models: type filtering and name search.
Use the type filter dropdown to show only checkpoints, LoRAs, embeddings, ControlNet models, VAEs, or other types. This is the fastest way to find a specific category of model.
Type a model name into the search bar to filter the list. The search matches against the model's display name and filename.
flux1-devdreamshaperWhen viewing an asset's detail page, the workflow section lists every model used in that generation. Each model name is a link that navigates directly to /app/models/[hash]. This is the quickest way to jump from a specific output to the model that produced it.
Every model's metadata is fully editable from the Overview tab. Add descriptions, apply tags, and write notes to organise your model collection.
Click any editable field on the Overview tab to modify it. Fields include description, display name, and notes. Changes are saved immediately via the PATCH API.
Add tags to categorise models by use case, quality, or any taxonomy that suits your workflow. Type a tag name and press Enter to apply it.
All changes are persisted as Data Vault satellite records. Each edit creates a new row, providing a complete, append-only audit trail.
Model metadata updates are sent via PATCH /api/v1/models/[hash]. The request body accepts description, tags, and notes fields. All writes are tenant-scoped and require authentication.