I already have a working ComfyUI-based image-generation setup, but it needs to run reliably at scale. My priority is to harden the existing pipeline so it can handle large batch jobs, quick regenerations, and seamless switching between several pre-trained models without crashing or slowing to a crawl. Here’s what I need from you: • Diagnose and eliminate the crashes and bottlenecks that appear during high-volume or repeat runs. • Optimise GPU/CPU memory usage so multiple models can be loaded, swapped, or unloaded cleanly. • Script or extend ComfyUI workflows for fully automated batch generation and one-click regeneration, tying in any helper utilities you find useful (Python, Bash, Docker, etc.). • Document the final setup clearly enough that I can reproduce it on another machine in a few minutes. No model training, fine-tuning, or research is required—this is purely an execution and automation task on top of the models I already have. If you’re comfortable reading existing node graphs, writing helper scripts, and squeezing maximum stability out of ComfyUI, I’d love to hand this over to you.