What is figmirror?

FigMirror is a Python tool that takes a figure from a research paper as a style reference and then redraws your own data to look like it belongs in that same paper. The idea is to eliminate the time researchers spend hand-tuning chart colors, fonts, spacing, and layout to match a target style. You supply an image of the reference figure and your data, and the tool produces both an editable Python script and a print-ready PDF of the new figure. Under the hood, FigMirror runs what it calls a Drawer and Reviewer loop. The Drawer generates a candidate chart using a charting library called matplotlib. The Reviewer then compares the candidate visually against the reference image, lists what needs to change, and notes which elements are already correct and should not drift further. This loop repeats until the output converges. The tool also maintains a running list of preserved style choices so that later iterations do not undo earlier progress. For three-dimensional figures, the system adds extra steps to handle camera angle, scale, lighting, and surface properties specific to 3D charts. The README describes a technique called Grounded Measurement, which lets the AI model read pixel coordinates and colors directly from the reference image and turn those readings into concrete checks during the revision process. You can run FigMirror in two ways. The web interface runs locally in a browser and shows upload controls, iteration history, and a preview panel. The skill mode installs FigMirror as a plugin inside Claude Code or Codex, letting you ask the AI assistant directly to mirror a figure's style. A gallery of 139 reference figures across 25 chart families is hosted online for users who do not have a specific paper figure on hand. The project is open to contributions, with the README pointing to areas like adding showcase examples, improving the reviewer loop, and writing regression tests. The codebase is written in Python and the install script handles dependencies.