Where to Put Safetensors in ComfyUI Models Folder

Put .safetensors files somewhere under ComfyUI/models/, but choose the subfolder by model type, not by the .safetensors extension. Checkpoints go in models/checkpoints/, LoRAs in models/loras/, VAE in models/vae/, Flux diffusion models usually in models/diffusion_models/, and text encoders in models/text_encoders/ or models/clip/.

If a file is in the wrong folder, ComfyUI may start normally but the dropdown is empty or the workflow fails validation.

Quick folder map

After moving files, restart ComfyUI or refresh the model list from the relevant dropdown.

If you downloaded a model and cannot find it

Manual downloads do not all land in one folder. A browser download usually stays in your downloads folder until you move it; a launcher, workflow helper, or custom node may place files under the active ComfyUI models/ folder, a Hugging Face cache, or a node-specific folder. If the file is not visible in a dropdown, first confirm you are editing the same ComfyUI install that started the server.

When a filename includes a quantization tag such as q8, fp8, fp16, or gguf, do not assume it belongs in models/checkpoints/. Those tags describe precision or format, not the loader. Use the workflow node name: checkpoints → checkpoints, text encoders → text_encoders/clip, VAE → vae, diffusion/GGUF models → diffusion_models or the legacy unet path.

If you fixed missing nodes but models are still missing

Missing nodes and missing model files are separate repair stages. After red nodes register, the workflow may still ask for model files — that does not mean the node repair failed; it means the workflow can now run far enough to reveal the next missing asset.

Common file examples

When a filename includes a model family such as Flux or Qwen, do not assume it belongs in models/checkpoints/. Check the loader node first.

Fast diagnosis

Workflow import checklist

1. Confirm red/missing nodes are fixed first.
2. Read the loader node name: checkpoint, LoRA, VAE, CLIP, ControlNet, diffusion model, GGUF, or upscaler.
3. Put the file in the folder for that loader, not the folder suggested by the extension.
4. Confirm you are moving files into the ComfyUI install that is actually running.
5. Restart ComfyUI or refresh the model list before deciding the file is missing.

If the next error changes from "model missing" to a Python import or plugin error, switch to No module named errors or plugin import failed.

Model types that share the .safetensors extension

A full checkpoint, a LoRA, a VAE, a ControlNet model, a Flux diffusion model, and a text encoder can all end with .safetensors but belong in different folders. Sorting by extension alone eventually puts the right file in the wrong folder.

• Checkpoints (realvisxl.safetensors, sd_xl_base_1.0.safetensors) → models/checkpoints/, use Load Checkpoint.
• LoRA → models/loras/, use Load LoRA with a compatible base model (SD 1.5 LoRA with an SD 1.5 checkpoint, SDXL with SDXL).
• VAE → models/vae/, use Load VAE. A wrong VAE can cause color/decode problems.
• ControlNet → models/controlnet/. Must match the model family (SD 1.5 and SDXL ControlNet are not interchangeable).
• Flux/UNet/diffusion models → models/diffusion_models/; some older workflows use models/unet/. ComfyUI maps legacy unet to diffusion_models, so both appear in guides.
• GGUF models → usually models/diffusion_models/ (some ComfyUI-GGUF examples use models/unet/), loaded with Unet Loader (GGUF). See ComfyUI GGUF.
• Text encoders / CLIP → models/text_encoders/ or models/clip/; ComfyUI supports both.

Why the dropdown is still empty

1. Check the extension — .safetensors is supported, but a partially downloaded file may not be.
2. Refresh the dropdown or restart ComfyUI.
3. Make sure the file is not inside a zip or browser temp folder.
4. Confirm you are editing the same ComfyUI folder you are running.
5. Check extra_model_paths.yaml if models live outside the ComfyUI folder.
6. If ComfyUI crashes at startup or logs environment errors while loading models, see Torch Not Compiled With CUDA Enabled.

What not to do

• putting every .safetensors file in models/checkpoints
• using Load Checkpoint for a Flux diffusion model
• putting LoRA files next to full checkpoints
• editing one ComfyUI install while launching a different one
• assuming a browser refresh is enough after moving many models

Portable package example

ComfyUI_windows_portable/
  ComfyUI/
    models/
      checkpoints/
      loras/
      vae/
      controlnet/
      diffusion_models/

If your launch .bat starts a different ComfyUI folder, the model will not show up. Verify the folder shown in the running terminal or launcher settings.

FAQ

Where do .safetensors files go in ComfyUI?

Somewhere under ComfyUI/models/, but the exact folder depends on whether the file is a checkpoint, LoRA, VAE, ControlNet model, Flux diffusion model, or text encoder.

Can I put all .safetensors files in models/checkpoints?

No. That is the most common mistake — many .safetensors files are not checkpoints.

Where do Flux .safetensors models go?

Usually ComfyUI/models/diffusion_models/ or the legacy models/unet/ path, loaded with a diffusion model or UNet loader.

Where do Q8 or FP8 models go?

Q8, FP8, and FP16 describe precision, not the folder. For a Flux diffusion or GGUF model, check models/diffusion_models/ or models/unet/. For clip_l, t5xxl, Qwen, or another text encoder, use models/text_encoders/ or models/clip/.

Why is my .safetensors file not showing up?

Usually because it is in the wrong folder, you are looking at the wrong loader, or you edited a different ComfyUI install than the one running.

How do I install a .safetensors model in ComfyUI?

There is no installer — installing a safetensors model means moving the file into the matching ComfyUI/models/ subfolder (checkpoints, loras, vae, controlnet, diffusion_models, text_encoders, or upscale_models), then restarting ComfyUI or refreshing the dropdown. The loader node determines the folder, not the file name.

How do I open or use a .safetensors file in ComfyUI?

You do not open the file directly. Add a loader node (Load Checkpoint, Load LoRA, Load VAE, Load ControlNet Model, UNet/Diffusion model loader, or Upscale Model Loader) in your workflow and pick the file from its dropdown. If the file does not appear, it is in the wrong folder or ComfyUI has not refreshed the model list.

Where is the safetensors folder in ComfyUI?

There is no single "safetensors folder" — the location is ComfyUI/models/ and the exact subfolder depends on the model type: checkpoints/, loras/, vae/, controlnet/, diffusion_models/, text_encoders/, embeddings/, or upscale_models/. On the Windows portable package the path is ComfyUI_windows_portable\ComfyUI\models\.

Related Guides

• Download Models
• ComfyUI GGUF
• MiniMax H3 Guide: T2V, I2V & Reference-to-Video
• SageAttention is not new enough version — LTX-2
• Install Custom Nodes
• ComfyUI Common Issues
• ComfyUI Torch Not Compiled With CUDA Enabled Fix

Source References

• Official ComfyUI model troubleshooting
• Official ComfyUI models concept
• ComfyUI folder path source