VILA-Lab/FigMirror

An Automated AI Agent Tool for Plotting Your Data in Any Paper's Figure Style.

What it solves

FigMirror eliminates the tedious process of manually editing plotting code to match the visual style of a specific academic paper. It allows users to provide a reference figure and their own data to automatically generate a polished, camera-ready PDF and an editable matplotlib script that mirrors the target style.

How it works

The system employs an agentic loop consisting of a Drawer and a Reviewer:

  1. Drawer Agent: Renders a candidate figure using "Grounded Measurement," which leverages foundation models to return x/y coordinates for visual targets and convert them into executable checks (e.g., checking pixel colors).
  2. Reviewer Agent: Performs a visual audit by comparing the candidate to the reference image. It provides structured feedback, including bounding boxes, a revision checklist, and a "preserve list" to prevent style drift across iterations.
  3. Orchestrator: Manages the delegation between agents and converts reviewer feedback into annotated images and notes for the next drawing pass.

For 3D figures, the system uses geometry-aware prompting to handle camera angles, lighting, and scale. An Aesthetic Lib provides fallback style rules when agents disagree or confidence is low.

Who it’s for

Researchers and data scientists who need to create high-quality, publication-ready figures that adhere to specific stylistic guidelines or match the visual identity of a particular paper family.

Highlights

  • Agentic Iteration: Uses a role-separated Drawer/Reviewer loop for continuous visual refinement.
  • Grounded Measurement: Uses coordinate-based visual targets and code-based pixel checks for precision.
  • ** uma Editable Output**: Generates full matplotlib scripts rather than just static images.
  • Multi-Interface Support: Available as a local Web UI or as skills for Claude Code and Codex.
  • 3D Support: Includes specialized prompting for 3D composition and geometry.

Related

  • Project
    llmsresearch/paperbanana

    An agentic framework that automates the generation of publication-quality academic diagrams and statistical plots from text descriptions and data.

  • Project
    LiteReality/LiteReality-Agent

    LiteReality‑Agent is an open‑source toolkit that converts a phone‑captured RGB‑D scan (via the LiteReality iOS app) into a fully‑articulated, graphics‑ready indoor 3D scene. An LLM‑driven agent coordinates vision models (TRELLIS, GroundingDINO) and Blender to generate, place, and polish objects, producing GLB/BLEND assets usable in game engines, AR/VR, or robotics. It runs on macOS or Linux, can use Modal’s cloud GPUs or a local ≥24 GB GPU, and requires API keys for image generation and the LLM.

  • Project
    amagine-ai/Amagine3D

    An AI-powered 3D design layer that generates editable parametric CAD models for hardware enclosures and assemblies from text and image requirements.

  • Project
    CRui5in/paper-ppt-agent

    An AI-powered tool that transforms academic papers (PDF/TeX) into editable PowerPoint presentations using a multi-agent pipeline for content extraction and visual design.

  • Project
    jihe520/MathModelAgent

    An AI agent framework designed to automate mathematical modeling competitions, from problem analysis and coding to generating submission-ready PDF papers.