How to illustrate medical notes with AI: a two-pass workflow that adds diagrams without slowing your note-taking down

Why diagrams in medical notes are worth the effort

Dual-coding theory, first formalised by Allan Paivio in 1991 and replicated repeatedly in medical education, says you remember a concept better when it is encoded as both words and a picture. Anatomy, embryology, pathology, pharmacology and surgery all lean heavily on spatial reasoning. A labelled diagram is not decoration; it is a second memory trace, and at exam time it is often the trace that fires first.

The problem has always been time. Drawing a clean diagram by hand during a lecture costs you the lecture. Drawing it afterwards from memory costs you twenty minutes and you usually skip it. AI illustration collapses that twenty minutes into under a minute, which is the only reason this workflow is realistic for a full timetable.

The principle: capture rough, render later

Do not try to draw a clean diagram during a lecture. You will either miss what is being said or end up with a half-decent diagram and half-decent notes. Sketch fast, label minimally with arrows and single words, and trust that you will render it later. The AI step is what makes 'render later' viable for the first time.

The two-pass workflow, step by step

Pass 1: during the lecture (zero overhead)

• Sketch any structure that comes up. Stick figures of anatomy are fine; correctness of the labels matters more than the lines.
• Label one or two key points with single words: 'CN VII', 'left coronary', 'B cell zone'.
• Mark the page with a star or a margin tag so you can find it again in the evening.
• Do not stop the note flow to make it pretty. The pretty version is generated, not drawn.

Pass 2: that evening (5 to 10 minutes for the whole day)

1. Photograph the page or import your tablet sketches into a sketch-first illustration tool. Angiosome is built for this; see ai-medical-illustration for a full tool comparison.
2. Tell the tool what the structure is in one line: 'sigmoid colon, blood supply, posterior view'. Context sets the anatomical conventions.
3. Generate. Review for one minute against your textbook or an atlas such as Gray's or Netter's.
4. Save the labelled image back into your notes. Digital notes: paste in. Paper notes: print and stick in, or move that page to digital.
5. Tag the file with the topic so you can find it later for flashcards and slides.

Choosing the right AI tool (and why sketch-first matters)

Text-to-image models like ChatGPT's image generator, Midjourney and Stable Diffusion will draw something anatomical from a prompt, but they hallucinate confidently. Ask for a brachial plexus and you may get five trunks instead of three, or a circle of Willis with the wrong communicating arteries. They were trained on internet imagery, not anatomy textbooks, and they do not know when the picture is wrong.

Sketch-first tools take your drawing as the structural input and render it in a chosen style. Because the topology comes from your hand, the anatomy stays correct as long as your sketch is correct. This is the entire reason a one-minute sketch is worth more than a perfectly written prompt: the sketch carries the spatial truth.

Where this workflow shines

• Anatomy modules: you will generate dozens of diagrams a week, and the per-diagram time matters more than anywhere else.
• Pathology: a marked-up histology slide or a tumour-spread diagram is more useful than three paragraphs of prose.
• Pharmacology: receptor diagrams, drug-action cartoons, second-messenger pathways and mechanism flowcharts.
• Surgical placements: sketch the procedure as it is done, render the clean version at the end of the day. See ai-medical-illustration for surgical-sketch examples.
• Embryology: rotational sequences and folding diagrams where text descriptions collapse without a picture.
• Radiology: annotate a labelled overlay on a teaching scan (never a real patient scan without anonymisation).

What to do with the rendered images (or it was not worth generating)

A rendered diagram is worth the minute you spent on it only if you reuse it. The simple rule: every image should appear in three places. Drop it into Anki as an image-occlusion card (see anki-ai-workflow-for-med-school for the full pipeline). Drop it into your study guide for the module. Use it as the diagram on a case-presentation slide or audit poster. The same image, in three contexts, is how anatomy actually sticks.

1. Place 1: in the day's notes, captioned with the topic and date.
2. Place 2: in an Anki deck as an image-occlusion card with three to five labels masked.
3. Place 3: in a topic-level slide deck or your finals revision folder, ready for case presentation or teaching.

Safety, plagiarism and academic integrity

Three rules cover almost everything that goes wrong. Never upload identifiable patient data: no scans with hospital headers, no clinical photographs, no operative notes with names. Anonymise everything. Never paste in copyrighted textbook figures as your sketch input. Always disclose AI-generated illustrations in formal submissions if your school's policy requires it; most UK and US schools now do.

On the disclosure point, the General Medical Council's 2024 guidance on generative AI in medical education is the clearest reference for UK students. US students should check the AAMC's emerging guidance and their own school's honour code. When in doubt, declare.

What this workflow does not fix

• Bad note structure. AI illustration is the polish; you still need the bones of the notes to be sound.
• Missing concepts. The model cannot add knowledge you did not capture. Pair with NotebookLM (see notebooklm-for-medical-school) for filling lecture gaps.
• Memorisation. Pretty diagrams do not memorise themselves. Spaced repetition does.
• Examiner-style precision. If your school marks diagrams against a strict labelling rubric, hand-label the final image yourself.