Configurations, benchmarks, lessons learned, setup notes, failures, comparisons, hardware and software observations. No padding, no advocacy — just what happened.
Three ways to run ComfyUI — a laptop RTX 4070 on Tailscale, a rented A5000 pod, and cloud API credits — compared on setup friction, cost per useful output, and whether each is worth using again.
The starting plan for the AI video generation experiment — which tools to test, how cost gets weighed against output quality, and what stays free regardless.
Future notes will cover private infrastructure builds, local agent experiments, and whatever else turns out to be worth writing down.