# Draw figures with AI

Describe a diagram, preview editable TikZ, and refine it in your paper.

Describe the figure in the assistant chat, alongside the paper it belongs to. Oleafly can turn that brief into editable TikZ or PGFPlots and compile the figure on its own before insertion.

> Draw a three-stage pipeline: collect samples, estimate uncertainty, and compare predictions. Use the notation in methods.tex. Preview it before inserting it.

## Start with a brief

Say what the reader should understand, which labels matter, and where the figure will go. Mention relevant project files with `@`. If you already have a sketch, add it to the project and ask the assistant to use it as a reference.

## Preview and refine

The assistant can write figure source, compile a standalone preview, and use the resulting log to repair errors. A vision-capable model can also inspect a rendered preview. Ask for a change in the same conversation: shorter labels, a wider layout, or consistent notation.

These figure tools are available for LaTeX projects with an engine that supports isolated figure compilation. Choose a connected model in [AI setup](/docs/ai-setup/).

## Insert editable source

Once the preview reads well, ask the assistant to insert it with a caption and label. Review the proposed edit using your approval controls, then compile the complete paper to check how it fits.

The TikZ source stays in the project. You can edit it directly, ask for another revision, or open supported diagram elements in the [diagram composer](/docs/figures-diagrams/).

## Plot research data

Use [the analysis workflow](/docs/workflows/data-analysis/) when the figure depends on a dataset. Ask the assistant to save the script and the data selection behind the plot so you can reproduce it.

For a recurring figure workflow, invoke `/oleafly-figure-prep` from [Skills](/docs/skills/).