From Sketch to Publication

A Guide to AI-Assisted Scientific Visuals

Getting Started

Workflow Suggestion: To experience the full lifecycle of creating a visual, we recommend taking one primary figure (e.g., a data plot) through all the relevant stages: from generation ("Idea to Code"), to refinement ("Modify & Refine"), to final captioning. However, feel free to create new or different types of visuals (like a diagram) for any task where it makes more sense for your research.

Your Tools: A dataset, whiteboard/paper for sketching, example visuals, and your research draft.

Documentation: Create and keep a Google Doc open in the class vault to record your prompts, code, and analysis. Open Class Vault Folder.

The Forward Pass: From Idea to Python Code

Describe a desired visual and ask the AI to generate the corresponding Python code. Please choose at least one experiment to focus on. If you have time, feel free to try others.

The Reverse Pass: From Image to Code

Use multimodal LLMs to interpret a visual and generate the code to replicate it. Explore the tasks below to practice reverse-engineering visuals.

The Art of Modification: Iterative Refinement

Use the AI as a coding assistant to modify and build upon existing figures. Explore the different methods below for refining your visuals.

The Narrative: Generating Captions and Titles

A visual is only as good as its caption. Explore the different methods below for generating accurate and descriptive captions.

Reflection and Final Analysis

Review your notes, submit your best figure, and consolidate your learnings.

Final Submission

As the final step, please add your best figure from this exercise to our shared class presentation. Create a new slide, add your figure, and include your name.

Open Google Slides Presentation

Final Questions for Your Notes