How to Troubleshoot and Solve ComfyUI Issues

# Low VRAM mode (uses cpu for text encoder)
python main.py --lowvram

# CPU mode (very slow but works with any hardware, only use as absolute last resort)
python main.py --cpu

# Disable previews (saves VRAM and processing)
python main.py --preview-method none

# Use optimized attention mechanisms
python main.py --use-pytorch-cross-attention
python main.py --use-flash-attention

# Async weight offloading
python main.py --async-offload

# Reserve specific VRAM amount for OS (in GB)
python main.py --reserve-vram 2

# Disable smart memory management
python main.py --disable-smart-memory

# Use different caching strategies
python main.py --cache-none      # Less RAM usage, but slower
python main.py --cache-lru 10    # Cache 10 results, faster
python main.py --cache-classic   # Use the old style (aggressive) caching. 

# Check Python version (3.9+ required, 3.12 recommended)
python --version

# Use virtual environment (recommended)
python -m venv comfyui_env
source comfyui_env/bin/activate  # Linux/Mac
comfyui_env\Scripts\activate     # Windows

# Update pip first
python -m pip install --upgrade pip

# Install dependencies
pip install -r requirements.txt

# For NVIDIA GPUs (CUDA 13.0)
pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu130

# For AMD GPUs (Linux only - ROCm 7.2)
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm7.2

# For virtual environments with NVIDIA packages
export LD_LIBRARY_PATH=$VIRTUAL_ENV/lib/python3.12/site-packages/nvidia/nvjitlink/lib:$LD_LIBRARY_PATH

# For conda environments
export LD_LIBRARY_PATH=$CONDA_PREFIX/lib/python3.12/site-packages/nvidia/nvjitlink/lib:$LD_LIBRARY_PATH

# Or find your Python site-packages automatically
PYTHON_PATH=$(python -c "import site; print(site.getsitepackages()[0])")
export LD_LIBRARY_PATH=$PYTHON_PATH/nvidia/nvjitlink/lib:$LD_LIBRARY_PATH

# You may also need other NVIDIA libraries
export LD_LIBRARY_PATH=$PYTHON_PATH/nvidia/cuda_runtime/lib:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=$PYTHON_PATH/nvidia/cublas/lib:$LD_LIBRARY_PATH

# Check installed NVIDIA packages
python -c "import site; import os; nvidia_path=os.path.join(site.getsitepackages()[0], 'nvidia'); print('NVIDIA libs:', [d for d in os.listdir(nvidia_path) if os.path.isdir(os.path.join(nvidia_path, d))] if os.path.exists(nvidia_path) else 'Not found')"

# Find missing libraries that PyTorch needs
python -c "import torch; print(torch.__file__)"
ldd $(python -c "import torch; print(torch.__file__.replace('__init__.py', 'lib/libtorch_cuda.so'))")

# For virtual environments, add to activation script
echo 'export LD_LIBRARY_PATH=$VIRTUAL_ENV/lib/python*/site-packages/nvidia/nvjitlink/lib:$LD_LIBRARY_PATH' >> $VIRTUAL_ENV/bin/activate

# For conda environments
conda env config vars set LD_LIBRARY_PATH=$CONDA_PREFIX/lib/python*/site-packages/nvidia/nvjitlink/lib:$LD_LIBRARY_PATH

# For global bashrc (adjust Python version as needed)
echo 'export LD_LIBRARY_PATH=$(python -c "import site; print(site.getsitepackages()[0])")/nvidia/nvjitlink/lib:$LD_LIBRARY_PATH' >> ~/.bashrc

# Check current library cache
ldconfig -p | grep cuda
ldconfig -p | grep nccl

# If missing, add library paths (requires root)
sudo echo "/usr/local/cuda/lib64" > /etc/ld.so.conf.d/cuda.conf
sudo ldconfig

# Verbose library loading to see what's missing
LD_DEBUG=libs python main.py 2>&1 | grep "looking for"

# Check PyTorch CUDA availability
python -c "import torch; print('CUDA available:', torch.cuda.is_available()); print('CUDA version:', torch.version.cuda)"

# After updating ComfyUI via Git, update frontend dependencies
pip install -r requirements.txt

# First uninstall torch
pip uninstall torch

# Install stable PyTorch with CUDA 13.0
pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu130

# For nightly builds (might have performance improvements)
pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu132

# Verify CUDA support
python -c "import torch; print(torch.cuda.is_available())"

# Check if GPU is visible
nvidia-smi

# Check driver version and CUDA compatibility
nvidia-smi --query-gpu=driver_version --format=csv

# Install stable ROCm PyTorch (7.2 at the time of writing)
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm7.2

# For nightly builds (ROCm 7.2 at the time of writing, which might have performance improvements)
pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/rocm7.2

# For RDNA2 or older (6700, 6600)
HSA_OVERRIDE_GFX_VERSION=10.3.0 python main.py

# For RDNA3 cards (7600)
HSA_OVERRIDE_GFX_VERSION=11.0.0 python main.py

# Enable experimental memory efficient attention (no longer necessary with PyTorch 2.4)
TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1 python main.py --use-pytorch-cross-attention

# Enable tunable operations (slow first run, but faster subsequent runs)
PYTORCH_TUNABLEOP_ENABLED=1 python main.py

# Install PyTorch nightly for Apple Silicon
# Follow Apple's guide: https://developer.apple.com/metal/pytorch/

# Check MPS availability
python -c "import torch; print(torch.backends.mps.is_available())"

# Launch ComfyUI
python main.py

# Force CPU mode
python main.py --cpu

# With memory optimization
python main.py --force-fp16 --cpu

# Install PyTorch nightly with XPU support
pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/xpu

# Launch ComfyUI
python main.py

# For Intel Arc A-Series Graphics
conda install libuv
pip install torch==2.3.1.post0+cxx11.abi torchvision==0.18.1.post0+cxx11.abi torchaudio==2.3.1.post0+cxx11.abi intel-extension-for-pytorch==2.3.110.post0+xpu --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/