FREE GUIDE · NIRMAL
AI & Workflows3 min readUpdated recently
ComfyUI fixes
Stop fighting broken workflows. These 5 quick configuration tweaks resolve 90% of ComfyUI startup and generation bottlenecks.
[ STEP-BY-STEP PLAYBOOK ]
1. VRAM & Model Loading Flags
- [1]Launch with '--lowvram' or '--medvram-sdxl' depending on your GPU limit.
- [2]Enable fp8 weight quantization for SDXL and FLUX models to cut memory footprint in half without noticeable fidelity degradation.
2. Resolving Custom Node Dependency Hell
- [1]Use ComfyUI-Manager's 'Try Fix' feature to reinstall missing pip dependencies directly into python embedded environment.
- [2]When git conflicts arise, delete the node directory inside /custom_nodes/ and git clone the clean upstream repo.
3. Fast Preview & Latent Caching
- [1]Set preview method to 'taesd' (Tiny AutoEncoder) for near-instant 0-overhead step previews during generation.
Shared by Nirmal · Creator, Designer & Builder