Where ComfyUI files live
The folders of ComfyUI: where to put each kind of model, where pictures come from and where results go.
The main folders
Inside the ComfyUI folder (for ComfyUI Desktop, inside the folder you chose at install):
| Folder | What lives there |
|---|---|
models/ | All model files, sorted by kind (below). |
input/ | Pictures, videos and sounds that Load Image, Load Video and Load Audio list. An upload in the editor lands here. |
output/ | What Save Image, Save Video and other output nodes write. |
temp/ | What Preview Image shows; cleaned up by ComfyUI. |
custom_nodes/ | Installed node packs, one folder each. |
user/ | Your settings and saved workflows (user/default/workflows). |
Model folders
| Folder | Files | Read by |
|---|---|---|
models/checkpoints | Whole checkpoints (SD 1.5, SDXL, Pony, Illustrious, some others) | Load Checkpoint |
models/diffusion_models (older name unet) | Diffusion models alone (Flux, Qwen-Image, Z-Image, Wan, HunyuanVideo) | Load Diffusion Model |
models/text_encoders (older name clip) | Text encoders (clip_l, t5xxl, umt5_xxl, Qwen) | Load CLIP, DualCLIPLoader |
models/vae | VAE files | Load VAE |
models/loras | LoRA files | Load LoRA and LoraLoaderModelOnly |
models/controlnet | ControlNet models | Load ControlNet Model |
models/upscale_models | ESRGAN-style upscalers | Load Upscale Model |
models/clip_vision | Image encoders (clip_vision_h and others) | Load CLIP Vision |
models/embeddings | Textual inversions | embedding:name in a prompt |
models/style_models, models/audio_encoders and others | Special-purpose models | Their own loader nodes |
Packs often add folders of their own (for example models/detection for pose models, models/wav2vec2 for speech models, or a GGUF folder), and their README says which.
Subfolders are fine: a file in models/loras/wan/ appears in the list as wan/name.safetensors.
To keep models elsewhere (a second disk, or shared with another program), ComfyUI reads extra_model_paths.yaml in its folder (ComfyUI Desktop sets this in its own settings). Restart ComfyUI after changing it.
Typical mistakes
- A file in the wrong folder: it does not appear in the loader's list. Move it and refresh (press R in the ComfyUI editor) or restart.
- A model folder named by another program (
Stable-diffusion,Lora): ComfyUI does not look there unlessextra_model_paths.yamlsays so. - Expecting files in
outputfrom Preview Image: preview files go totemp.
How the app helps
The app reads the lists of model files from your ComfyUI; after you add files, choose Sync ▸ Sync now. The Models tab of the library shows them, and Check names a model a graph asks for that is not on disk, with its folder. The app never downloads or moves files.