Common Issues

Installation Issues

"CUDA is not available" or "Torch not compiled with CUDA"

Cause: PyTorch was installed without CUDA support, or the CUDA version doesn't match your NVIDIA driver.

Fix:

1. Check your NVIDIA driver version: run nvidia-smi in Command Prompt
2. Reinstall PyTorch with the correct CUDA version:

pip uninstall torch torchvision torchaudio
pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu126

1. If you're using the portable package, make sure you downloaded the NVIDIA version and are running run_nvidia_gpu.bat

ComfyUI Desktop shows "Unsupported device"

Cause: ComfyUI Desktop on Windows requires an NVIDIA GPU with CUDA. AMD and Intel GPUs are not supported in the Desktop version.

Fix: Use the Portable Package or Manual Install instead.

Installer or app blocked by antivirus

Cause: Windows Defender or third-party antivirus flagging ComfyUI as suspicious.

Fix:

1. Add ComfyUI's install directory to your antivirus exception list
2. For Windows Defender: Settings → Privacy & Security → Virus & Threat Protection → Manage Settings → Exclusions
3. Re-download and try again

7-Zip extraction fails

Cause: The downloaded file is blocked by Windows or the path is too long.

Fix:

1. Right-click the .7z file → Properties → check Unblock → Apply
2. Extract to a short path like D:\ComfyUI instead of a deeply nested folder
3. Make sure you're using 7-Zip, not Windows built-in zip

"Permission denied" or package install failures

Cause: Running ComfyUI or its installer as Administrator, or installing to a system-protected folder.

Fix:

1. Never run ComfyUI as Administrator — this causes Python package permission conflicts
2. Do not install to C:\Program Files, C:\Windows, or C:\ root
3. Use a non-system path like D:\ComfyUI or C:\Users\YourName\ComfyUI

Windows Long Path Limit

Cause: Windows has a default 260-character path limit. Deep folder structures in custom nodes can exceed this.

Fix: Enable long paths in Windows 10/11:

1. Press Win + R, type regedit, press Enter
2. Navigate to HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\FileSystem
3. Set LongPathsEnabled to 1
4. Restart your computer

Or run this in PowerShell (as Administrator):

New-ItemProperty -Path "HKLM:\SYSTEM\CurrentControlSet\Control\FileSystem" -Name "LongPathsEnabled" -Value 1 -PropertyType DWORD -Force

---

Runtime Issues

CUDA Out of Memory

Cause: Your GPU doesn't have enough VRAM for the model or resolution you're using. See System Requirements for VRAM guidance per model type.

Fix (try in order):

1. Close other GPU applications (browsers, games, other AI tools)
2. Reduce the image resolution (e.g., 512×512 instead of 1024×1024)
3. Add --lowvram flag to your launch command
4. Use GGUF quantized models (see Download Models)
5. For video models: reduce frame count and resolution

Red Nodes in Workflow

Cause: The workflow uses custom nodes that you haven't installed.

Fix:

1. Install ComfyUI Manager
2. In Manager, click Install Missing Custom Nodes
3. Restart ComfyUI

"No checkpoint found" / Empty model dropdown

Cause: No model files in the checkpoints folder, or files are in the wrong location.

Fix:

1. Download a model (see Download Models)
2. Place the .safetensors file in ComfyUI/models/checkpoints/
3. Click Refresh in the model dropdown, or restart ComfyUI

Browser shows blank page or title only

Cause: Browser incompatibility.

Fix: Use the latest version of Google Chrome. Some browsers (especially older Edge or Firefox versions) may not render the ComfyUI interface correctly.

Generation is extremely slow

Cause: ComfyUI may be running on CPU instead of GPU.

Fix:

1. Check the console output — it should show your GPU name on startup
2. If using portable: make sure you're running run_nvidia_gpu.bat, not run_cpu.bat
3. If using manual install: verify PyTorch has CUDA support:

python -c "import torch; print(torch.cuda.is_available())"

Should print True. If it prints False, reinstall PyTorch with CUDA.

---

GPU-Specific Issues

RTX 50 Series (5070 Ti / 5080 / 5090)

See GPU Compatibility for the full driver-CUDA-PyTorch matrix. The RTX 50 series requires CUDA 12.8+ and specific PyTorch builds:

1. Update your NVIDIA driver to the latest version
2. Install PyTorch with CUDA 13.0:

pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu130

1. For the portable package, use the latest release with CUDA 13.0 support

Common 50-series issues:

• SageAttention compilation errors — requires updated build tools
• Nunchaku plugin failures — check for 50-series compatible versions
• xformers crashes — use PyTorch's built-in attention (--use-pytorch-cross-attention)

AMD GPUs on Windows

See GPU Compatibility for full AMD support details. AMD support on Windows uses DirectML, which has limitations:

• Some custom nodes don't support DirectML
• Performance is lower than CUDA
• Use the portable package with run_cpu.bat and add --directml flag

---

Network Issues

pip install hangs or times out

Cause: Network restrictions, firewall, or proxy blocking Python package downloads.

Fix:

1. Try a mirror:

pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple

1. If behind a proxy, configure pip:

pip install --proxy http://your-proxy:port -r requirements.txt

1. In ComfyUI Desktop: change the mirror settings in the setup wizard

Git clone fails

Cause: GitHub is blocked or rate-limited in your region.

Fix: Use a mirror or download the ZIP directly from the GitHub release page.

---

Not Finding Your Issue?

Try the Troubleshooting Decision Tree for a systematic diagnosis approach. For complex multi-plugin environments, see Dependency Conflicts and Workflow Environment Setup.

Still Stuck?

If none of the above solutions work, you can book a remote fix session with our experts. We connect via screen share and diagnose your specific setup — most issues are resolved in under 30 minutes.