google-research/papervizagent — reverse-engineered prompt

Build me a Python tool called PaperVizAgent that helps researchers turn a paper method section and a figure caption into polished academic diagrams or plots.

I want a simple web demo where I can paste the method text, add the caption, choose settings like diagram or plot, number of candidates, aspect ratio, whether to use references, and how many critique rounds to run. The app should generate several candidate images in parallel, show the progress and evolution of each one, and let me download individual images or a zip.

Please make the workflow feel like a small creative team, with steps for finding reference examples, planning the figure, improving the academic style, generating the visual, and critiquing it for refinements. Also add a second tab where I can upload an existing image, describe changes, upscale it to 2K or 4K, and download the result.

Include config support for API keys, a command line option for batch experiments, and basic visualization pages for reviewing pipeline results.