Models Manager

Models Manager

Models are the foundation everything else builds on — the checkpoint that defines your image's character, the LoRA that adds a specific style, the ControlNet that guides composition. The Models Manager gives you a single place to browse what you have, download what you need, and fix issues when a workflow can't find something it depends on.

Browsing Your Models

Open the Models workspace from the sidebar to see a grid of everything installed on your system. You can browse by type (checkpoints, LoRAs, VAEs, ControlNet models, etc.) using the taxonomy filters.

The grid view gives you a dense, scannable overview. Hover over any model to see the hover inspector — a quick-inspect overlay that shows the model's name, type, file size, and any enrichment metadata without leaving the grid.

Tip: If you have a large model collection, the taxonomy filters save a lot of scrolling. Start with the model type your workflow needs, then narrow from there.

Downloading Models

Sweet Tea supports downloading models directly from Civitai and Hugging Face with built-in progress tracking.

1. Open the download interface in the Models workspace.
2. Search or paste a URL for the model you want.
3. Start the download. Progress appears with a percentage bar and estimated time.
4. When it finishes, the model is automatically placed in the correct directory for its type.

You can cancel in-progress downloads if you realize you grabbed the wrong file. After download, verify the file placement before running a workflow that depends on it — some model types have specific path requirements.

Note: Large model downloads (10+ GB for some SDXL checkpoints) can take a while depending on your connection. Downloads run in the background, so you can keep working while they finish.

Fixing Missing Model Errors

The most common reason a Pipe won't run after import is a missing model. Here's how to resolve it:

1. Open the failing Pipe. The error message will tell you which model is missing and what type it is (checkpoint, LoRA, clip_vision, text_encoder, etc.).
2. Open the Models workspace and check if you have the model under a different name or path.
3. If you don't have it, download it using the steps above.
4. If you do have it but the path doesn't match, check the path configuration (below).
5. After installing or fixing the path, go back to the Pipe and try generating again.

Tip: Pay attention to the model type in the error message. A clip_vision model needs to be in the clip_vision directory, not mixed in with checkpoints. Sweet Tea's improved model-kind handling helps route you to the right fix location.

Path Configuration

Models are organized by type in your ComfyUI directory:

ComfyUI/
└── models/
    ├── checkpoints/
    ├── loras/
    ├── vae/
    ├── controlnet/
    ├── clip_vision/
    ├── text_encoders/
    └── ...

If you've moved your model directory or have models spread across multiple locations, verify the directory mapping in Settings. Path mismatches are the most frequent cause of "model exists but workflow can't find it" issues.

Make sure:

• The model directory mapping points to the real location on disk
• The ComfyUI runtime can read that directory
• Model file names match what the workflow expects (or are remapped in the Pipe's dependencies)

Model Enrichment

Enrichment jobs add metadata to your installed models — pulling in descriptions, preview images, model family information, and other context from the original source. This turns a bare file listing into an informative catalog.

Enrichment runs in the background and helps you:

• Distinguish between similar models with ambiguous filenames
• Verify the model family and intended use case before first use
• Reduce the chance of accidentally using the wrong model in a complex pipeline

Note: Enrichment metadata comes from external sources and may not be available for every model. Models you've created yourself or obtained through non-standard channels will need manual annotation.

Next: Extension Manager