Why is my ComfyUI model missing from the dropdown?

Match the file, folder and loader

A missing ComfyUI model dropdown usually calls for three checks: the running installation's search paths, the file's role and the loader already connected to the workflow. Moving every .safetensors file into checkpoints can create another mismatch. The extension alone does not identify a checkpoint, LoRA or separate diffusion model.

Use your existing installation, the original workflow and its creator's dependency list. Keep a copy of the workflow before changing it. These instructions were checked against current documentation and core source on 12 October 2026; we have not installed ComfyUI, downloaded weights or generated an image for this answer.

1. Identify the missing item

Write down the exact error, loader name, requested filename and model family. Is the loader present but its file is missing, or is the node itself unavailable? Those are different failures; use our missing-node answer for the latter.

Open the workflow's model links or README. Compare the requested variant with your actual download. A workflow JSON is a recipe; it does not contain the weights. A LoRA also needs the compatible base model. Do not rename a different file to satisfy the warning.

The official model overview explains the file and loader relationship. For a known beginner reference, our first ComfyUI workflow uses one specified SD1.5 checkpoint and built-in nodes. A Flux or video workflow may need several separate files and a different graph.

2. Open the directories used by this installation

Current Comfy Desktop: open the top-left menu, then Desktop Settings → Storage. Under Shared Models, use the folder icon to open a listed directory. The directory marked Primary is the default destination for new model downloads. Also check the running instance's Manage → Storage overrides if it excludes shared directories or adds its own. These controls are documented in Desktop Settings.

Legacy Desktop: older single-install versions have different menus and locations. The model documentation describes a capsule menu with a Storage tab. Use the actual path shown in your app rather than copying another person's AppData path. The Desktop FAQ distinguishes the legacy app from today's multi-install manager.

Windows Portable: open the extracted package you actually launch. Its model directory is inside the ComfyUI folder, for example ComfyUI_windows_portable/ComfyUI/models/. Check the equivalent location inside your AMD or Intel package if its outer name differs. See the Portable folder structure.

Manual installation: use the models directory in the running checkout, including any configured extra paths. Having two installations means the nearest folder in File Explorer is not necessarily the one this process searches.

3. Put each component in the matching category

The following are common built-in loader pairings relative to a recognised model root. Follow the specific workflow's instructions for custom loaders.

On a small screen, scroll the table sideways to see all three columns.

The checkpoint node reference, diffusion loader reference, CLIP reference and VAE reference describe these roles. We also checked the core loader implementation, because the node reference pages disclose that they were AI-generated.

Current folder-path source also searches legacy unet for diffusion models and clip for text encoders. These aliases do not make a LoRA into a checkpoint. A file appearing in a dropdown is only a discovery check: its contents still need to be compatible with the loader and graph.

Confirm the download finished, its full filename matches the intended variant and its size matches the publisher's file listing. If the publisher supplies a checksum, compare that too. Keep original filenames so your notes remain meaningful.

4. Add an existing model library without duplicating it

In current Desktop, use Storage → + Add Shared Directory, then check any per-instance override. If models disappeared after migration, the migration guide says existing weights stay in their old folder: the new installation adds that folder as a shared directory. Check the migrated instance's Manage → Storage → Model Directories before downloading again.

For Portable/manual installations, back up any existing extra_model_paths.yaml first. If none exists, copy extra_model_paths.yaml.example in the ComfyUI root and name the copy extra_model_paths.yaml. Do not leave the .example suffix or add .txt.

This original minimal example assumes you already have D:/AI/model-library/checkpoints containing suitable checkpoints. Replace the example base path with your real model-library root; do not paste it unchanged:

my_model_library:
  base_path: D:/AI/model-library
  checkpoints: checkpoints

Add this block alongside existing configuration, using spaces and a unique top-level name. It adds a checkpoint search path; it does not move files, fetch weights or configure the other categories. The upstream configuration example documents category-specific paths. Restart your ComfyUI process after saving. If startup fails immediately, restore the backed-up configuration and inspect the error before retrying. Desktop users should use their Storage controls rather than replacing generated configuration with this Portable example.

5. Refresh and select the file in the connected loader

For a file copied into an already recognised folder, focus the canvas and press R to refresh node definitions, then reopen the relevant dropdown. The default shortcut reference documents this action; customised keys can differ. If needed, save your work and restart the same installation.

Select the actual file in the loader already connected to your graph. Dragging a model from a library can create a separate node; an unconnected node does not replace the template's active loader. Check the wires and all companion-model selections before clicking Run once.

If it now lists the file but fails while loading, retain that new error. A format, unsupported model or missing encoder failure calls for the matching model instructions; an out-of-memory error calls for capacity/runtime checks. Repeated refreshes do not fix either. For custom-node formats such as GGUF, confirm the required project's loader and dependencies instead of forcing the file through Load Checkpoint.

6. Keep a recoverable result

When your workflow completes, save its output locally and reopen that file outside ComfyUI. Use Workflows → Export to keep a separate JSON, then Workflows → Open to check it reloads with the selected model filenames. The first-generation instructions document these workflow actions and saving an image from Save Image.

Keep the JSON, model source URLs, full filenames, directory notes and app/core versions together. Exporting a workflow does not bundle its models or make machine-specific paths portable. If generation still fails, export the diagnostic workflow and keep the error; do not label it working.

Would paying for hosting help?

Fixing local discovery needs no subscription. Comfy Cloud pricing describes hosted compute, runtime credits and custom-model importing on specified paid plans. A hosted template with the required model available can avoid maintaining local folders and hardware. Check model availability, supported nodes, actual charges and data handling before choosing it: payment does not repair an incompatible recipe or guarantee a successful output. No paid route was executed for this answer.