How to use AI to make a graphical abstract in minutes

A graphical abstract in a paper is one of the most useful tools for communicating the hard-earned results of research. It makes findings more memorable, accessible and engaging for everyone, from domain experts and policymakers to curious readers outside the field.

But creating graphical abstracts isn’t easy, and not everyone has access to a graphic designer, or is comfortable with design software. Artificial-intelligence tools can help fill the gap — if they are used conscientiously.

Three tips to avoid AI image mistakes in science

Pre-AI, when I designed graphical abstracts, I spent significant time working out the dos and don’ts, which tools to use and how to begin. Now, I can use AI to dictate ideas through its speech-to-text functions, vectorize sketches, optimize colour palettes for accessibility, find references, brainstorm ideas and build custom tools to suit an exact workflow.

But with greater convenience comes greater responsibility. An AI-assisted graphic has your name on it, not the AI’s. It is on you, the user, to ensure that the image is scientifically accurate and neither plagiarizes nor infringes on others’ work.

So, how to create quality graphical abstracts while integrating AI responsibly and effectively? Here are some of the lessons I’ve learnt.

Think before you start

First, check your target journal’s guidelines on the use of AI. Some journals prohibit AI-generated images and others require a clear declaration.

Assuming that AI use is allowed, your next step is to work out what you want to communicate. Are you illustrating a mechanism, a process, an experimental set-up, a workflow, a comparison or a timeline? For one of my papers, the key message was how enzymes called deubiquitinases drive obesity-related liver pathology, and where in the disease process enzyme inhibitors would halt the progression or development of disease¹. So, instead of mapping every individual molecular interaction, the graphical abstract — which in this case was created without AI — was designed to show how these enzymes contribute to disease progression using a single representative pathway rather than the full network of interactions identified in the literature. The goal was to provide enough context to make the mechanism intuitive and engaging, while leaving room for readers to explore the molecular details in the full study.

Then, think about your target readers. Are they clinicians, industry professionals, policymakers, students or domain experts? Knowing your audience helps you to determine the appropriate level of detail, terminology and visual complexity. Academic readers might prefer a biochemical focus, whereas physicians look at clinical implications, such as inflammation and insulin resistance. Make the abstract relevant to your audience by focusing on the part of the research that matters to them.

‘Good design takes mastery’: scientific illustrators sketch out AI’s future

Finally, look to the literature. How do similar papers present their graphical abstracts, both in content and in composition? Which ones work well, and why?

AI can help to augment these brainstorming steps. Tools such as ChatGPT (from OpenAI in San Francisco, California) and Claude (from Anthropic in San Francisco) can help to distil the study’s key message, identify relevant papers and let you better understand your readers. AI tools can even suggest pictorial representations of broad terms or concepts that you want to portray in the abstract, such as multiomics or systems-level interactions.

Communicate with design

What an AI tool creates depends on how clearly you communicate your idea. And for that, you need a well-written prompt. Make intentional decisions on such visual aspects as:

Colour: Aligning colour choices with psychology makes the abstract more intuitive. Red and orange draw the eye and are often perceived as negative or dangerous, so researchers frequently use ‘warm’ colours such as these to indicate key interactions, or harmful elements such as cancer cells. ‘Cool’ colours, by contrast, tend to signal normal or healthy conditions.

Celebrating Nature’s covers — past, present and future

But less is more, and too much colour (or too many shapes) can cause confusion. Try to create a limited, consistent and high-contrast palette.

Ask your AI tool: Which colour can I use to ensure colour accessibility on a given background? Or, what colours should I assign if I want to show a particular pathway in a specific cell type?

Composition: Do you prefer a horizontal or vertical flow, a cyclic loop or a forking pathway? Each directs the eye differently. A good composition guides the eye through the figure in the intended direction.

Ask your AI tool: Here is a graphical abstract for my research. Can you suggest alternative compositions? Can you recommend alternative text placement?