Errors & Fixes
Step-by-step guides to fix the most common ComfyUI errors.
Gemma 4 12B and LTX-2.5: diagnosing the projection error in ComfyUI
Your Gemma encoder loads, the prompt encodes, then LTX-2.5 dies with 'got 4 and 3'. We reproduced it on an RTX 3090 and traced it to one missing tensor.
ComfyUI "Import Failed" / Custom Node Errors: How to Actually Fix Them
Fix ModuleNotFoundError and 'Import Failed' errors in ComfyUI custom nodes. Install missing dependencies, resolve version conflicts, and fix Windows C++ build failures.
"Torch Not Compiled with CUDA Enabled" in ComfyUI: The Real Fix
Fix the 'Torch is not compiled with CUDA support' error in ComfyUI. Real cause: CPU-only PyTorch, wrong CUDA index, or a broken virtual environment.
CUDA Out of Memory in ComfyUI: How to Fix It (2026)
Fix the CUDA out of memory error in ComfyUI step by step. Memory flags, resolution, quantized models and VAE tiling for low-VRAM GPUs (4GB, 6GB, 8GB).