crisng95/flowboard ? reverse-engineered prompt

Build me a local only app called Flowboard for making AI product videos on an infinite canvas. I want to drag in reusable nodes for things like a model reference, a product reference, a composed image, a scene variation, and a video, then connect them so each step depends on the previous one. The main point is that I should only upload a model or product once, reuse it across the board, and click Generate on any image or video node to create results and regenerate variants without losing the graph.

It should be single user and run on my machine, with a Chrome extension that uses my logged in Google Flow session so generation works inside Flow. Please support auto written prompts based on upstream context and image understanding, using an installed CLI provider like Claude Code by default, with Gemini or Codex as options if available. The canvas should make it easy to see the whole campaign workflow from refs to images to videos. Look up current docs online if you need to, and wire up whatever is needed so it actually runs end to end.