Where Every Model File Goes

Folders

The definitive folder structure guide for ComfyUI

#Overview

Model folder confusion is the #3 most common problem on Reddit. This guide shows you exactly where every file type belongs.

One simple rule: Every model type has its own specific folder. Put files in the wrong place = ComfyUI won't see them.

Hardware Partner

Running these workflows? ComputeAtlas.ai helps you find the right GPU

Optimization is only half the battle. Get precise VRAM benchmarks and hardware recommendations tailored for ComfyUI.

Check GPU Prices →

[AD VISUAL: GPU BENCHMARKS]

#📁 Complete Folder Structure

comfyui-workflow.json

ComfyUI/
└── models/
    ├── checkpoints/          ← SD models (.safetensors, .ckpt)
    ├── loras/               ← LoRA files (.safetensors)
    ├── vae/                 ← VAE files (.safetensors, .pt)
    ├── controlnet/          ← ControlNet models
    ├── clip/                ← CLIP models
    ├── clip_vision/         ← CLIP Vision models
    ├── embeddings/          ← Textual Inversion embeddings (.pt, .safetensors)
    ├── unet/                ← UNet models (Flux, newer architectures)
    ├── upscale_models/      ← Upscaler models (ESRGAN, RealESRGAN)
    ├── ipadapter/           ← IPAdapter models
    ├── hypernetworks/       ← Hypernetwork files
    ├── style_models/        ← Style transfer models
    └── insightface/         ← Face restoration models

#🎯 Most Common Files & Where They Go

Checkpoints (Base Models)

File types: .safetensors, .ckpt

Examples:

→v1-5-pruned-emaonly.safetensors (SD 1.5)
→sd_xl_base_1.0.safetensors (SDXL)
→flux1-dev.safetensors (Flux)

Location:

comfyui-workflow.json

ComfyUI/models/checkpoints/

How to use: Load with the "Load Checkpoint" node in ComfyUI

LoRAs

File types: .safetensors, .pt

Examples:

→add_detail.safetensors
→anime_style_lora.safetensors
→photorealistic_face.safetensors

Location:

comfyui-workflow.json

ComfyUI/models/loras/

How to use: Load with "Load LoRA" node

⚠️ Critical: LoRAs MUST match your base model:

→SD 1.5 LoRA → SD 1.5 checkpoint
→SDXL LoRA → SDXL checkpoint
→Flux LoRA → Flux checkpoint

VAE Files

File types: .safetensors, .pt

Examples:

→vae-ft-mse-840000-ema-pruned.safetensors
→sdxl_vae.safetensors

Location:

comfyui-workflow.json

ComfyUI/models/vae/

How to use: Load with "VAE Loader" node → Connect to "VAE Decode"

When you need a VAE:

→Images look washed out, blurry, or have wrong colors
→Using SDXL without built-in VAE
→Workflow specifically requires external VAE

ControlNet Models

File types: .safetensors, .pth

Examples:

→control_v11p_sd15_openpose.pth
→diffusers_xl_canny_full.safetensors

Location:

comfyui-workflow.json

ComfyUI/models/controlnet/

How to use: Load with "Load ControlNet Model" node

Note: ControlNet models are also version-specific:

→SD 1.5 ControlNet → SD 1.5 checkpoint
→SDXL ControlNet → SDXL checkpoint

Embeddings (Textual Inversion)

File types: .pt, .safetensors

Examples:

→bad-hands-5.pt
→EasyNegative.safetensors
→negative_hand.pt

Location:

comfyui-workflow.json

ComfyUI/models/embeddings/

How to use: Just place files here. Use in prompts by filename (without extension):

comfyui-workflow.json

beautiful landscape, (bad-hands-5:0.8)

Upscale Models

File types: .pth, .safetensors

Examples:

→RealESRGAN_x4plus.pth
→4x-UltraSharp.pth
→ESRGAN_4x.pth

Location:

comfyui-workflow.json

ComfyUI/models/upscale_models/

How to use: Load with "Upscale Model Loader" node

IPAdapter Models

File types: .safetensors, .bin

Location:

comfyui-workflow.json

ComfyUI/models/ipadapter/

Typical structure:

comfyui-workflow.json

ipadapter/
├── ip-adapter_sd15.safetensors
├── ip-adapter-plus_sd15.safetensors
└── ip-adapter_sdxl.safetensors

#🔧 Organization Tips

Create Subfolders (Recommended)

ComfyUI supports subfolders. Organize like this:

comfyui-workflow.json

checkpoints/
├── SD15/
│   ├── realistic/
│   │   └── realisticVision_v51.safetensors
│   └── anime/
│       └── anything-v5.safetensors
└── SDXL/
    └── base/
        └── sd_xl_base_1.0.safetensors

loras/
├── SD15/
│   ├── style/
│   └── character/
└── SDXL/
    └── style/

Why?

→Easier to find files
→Prevents version mixing
→Better performance (fewer files to scan per folder)

#🚨 Common Folder Mistakes

❌ Wrong: Putting LoRAs in checkpoints folder

comfyui-workflow.json

checkpoints/
└── my_lora.safetensors  ← WRONG

Result: LoRA won't load, or will crash workflow

❌ Wrong: Mixing SD 1.5 and SDXL in same folder without subfolders

comfyui-workflow.json

checkpoints/
├── sd15_model.safetensors
└── sdxl_model.safetensors  ← Confusing

Better:

comfyui-workflow.json

checkpoints/
├── SD15/
│   └── sd15_model.safetensors
└── SDXL/
    └── sdxl_model.safetensors

❌ Wrong: Using OneDrive/Google Drive synced folders

Never put models in:

→OneDrive
→Google Drive
→Dropbox
→iCloud Drive

Why?

→Cloud sync locks files
→Causes loading errors
→Extremely slow model loading

#📥 How to Download Models

From Hugging Face

→Find the model page
→Click "Files and versions"
→Download .safetensors files
→Move to correct ComfyUI folder

Example URLs:

→SD 1.5: https://huggingface.co/runwayml/stable-diffusion-v1-5
→SDXL: https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0

From Civitai

→Find the model
→Click "Download"
→Choose .safetensors format (preferred over .ckpt)
→Move to correct folder

File size guide:

→SD 1.5 checkpoint: ~2-4 GB
→SDXL checkpoint: ~6-7 GB
→LoRA: 10-200 MB
→ControlNet: 700 MB - 2.5 GB

#🔍 Troubleshooting: "Model Not Found"

Model exists but ComfyUI doesn't see it?

Check:

→
File is in correct folder
- →Checkpoints in models/checkpoints/
- →LoRAs in models/loras/
→
File extension is correct
- →.safetensors or .ckpt (checkpoints)
- →.safetensors or .pt (LoRAs, VAE, embeddings)
→
File isn't corrupted
- →Check file size matches expected
- →Re-download if suspiciously small
→
Restart ComfyUI
- →Some folders need restart to refresh
- →Press Ctrl+C in console, restart
→
Check file permissions
- →File isn't "Read Only"
- →Folder isn't locked by cloud sync

#🎯 Model Recommendations by Use Case

For Photorealism

Checkpoint:

comfyui-workflow.json

models/checkpoints/
└── realisticVision_v51.safetensors

LoRAs:

comfyui-workflow.json

models/loras/
├── add_detail.safetensors
└── photorealistic_lighting.safetensors

For Anime/Illustration

Checkpoint:

comfyui-workflow.json

models/checkpoints/
└── anything-v5.safetensors

LoRAs:

comfyui-workflow.json

models/loras/
├── anime_lineart.safetensors
└── studio_ghibli_style.safetensors

For ControlNet Workflows

Must have:

comfyui-workflow.json

models/controlnet/
├── control_v11p_sd15_canny.pth
├── control_v11p_sd15_openpose.pth
└── control_v11p_sd15_depth.pth

#📊 Quick Reference Table

File Type	Extension	Folder	Average Size
SD 1.5 Checkpoint	`.safetensors`	`checkpoints/`	2-4 GB
SDXL Checkpoint	`.safetensors`	`checkpoints/`	6-7 GB
LoRA	`.safetensors`	`loras/`	10-200 MB
VAE	`.safetensors`	`vae/`	300-800 MB
ControlNet	`.pth`, `.safetensors`	`controlnet/`	700 MB - 2.5 GB
Embedding	`.pt`, `.safetensors`	`embeddings/`	5-100 KB
Upscaler	`.pth`	`upscale_models/`	15-65 MB

#💾 Managing Disk Space

Models take up massive space

Space-saving tips:

→
Delete unused checkpoints
- →One SD 1.5 + One SDXL is often enough
→
Use pruned/EMA versions
- →Smaller file size, same quality
- →Look for "pruned-emaonly" in filename
→
Store on external SSD
- →Symlink to external drive:
comfyui-workflow.json
mklink /D "C:\AI\ComfyUI\models\checkpoints" "E:\AI_Models\checkpoints"
→
Use .safetensors instead of .ckpt
- →Often smaller
- →Faster loading
- →More secure

#🆘 Still Can't Find Your Models?

→ Guide 5: Workflow Errors (Node not found issues)
→ Guide 9: Troubleshooting Assistant

Looking to optimize model loading speed? See Guide 7: Performance Optimization

Hardware Partner

Running these workflows? ComputeAtlas.ai helps you find the right GPU

Optimization is only half the battle. Get precise VRAM benchmarks and hardware recommendations tailored for ComfyUI.

Check GPU Prices →

[AD VISUAL: GPU BENCHMARKS]