Reconstruct scientific figures with AI
Rebuild authorised figures as editable SVG artwork, distinguish structural redraw from fidelity-led tracing and verify labels, layering and rendered differences.
Reconstruct scientific figures with AI around a clear priority: convenient editing or closer visual reproduction. The reconstruct-scientific-svg skill distinguishes a semantic redraw from tracing followed by correction. Both routes keep the SVG as source and require explicit checks of labels, layering and remaining raster content.
A high path count can preserve painted detail while making reshaping cumbersome. Native text, outlined glyphs and pixel-traced lettering are different deliverables. The workflow does not certify scientific or medical accuracy, and a numerical image comparison cannot replace content review.
What it does
- Rebuild publication figures with movable components
- Keep labels as genuine editable text
- Repair occlusion and back-to-front ordering
- Compare a reference with a matched-size render
- Export PNG or PDF derivatives from an SVG source
- Document required fonts and editing limitations
Install and use
- Supply the complete folder and an authorised readable reference to a file-and-image-capable agent. The full route needs Python, a renderer such as Inkscape and actual fonts; image metrics need NumPy and Pillow. PDF font inspection uses a separate tool.
- Try: “Reconstruct this three-stage experiment diagram. Keep labels as native text and each stage independently movable. Compare the output with the reference.” State whether editing or fidelity takes priority.
- Inventory shapes and labels, choose the reconstruction route and build deliberate layers. Remove traced lettering carefully before adding native text so unrelated artwork survives.
- Validate the SVG and inspect a same-size preview. Review boundaries, fonts and partial overlaps yourself. If exporting PDF, verify its actual text/font retention rather than assuming the SVG settings carried through.
FAQ
Source
Made by Yingqi-Han. Original skill: reconstruct-scientific-svg, published on GitHub under the MIT licence.
View on GitHub