Graphical Abstracts: What You Can Draft, What Journals May Reject

Graphical Abstracts: What You Can Draft, What Journals May Reject

A graphical abstract is a short visual summary of a study’s question, method, and main finding. It is an explanatory figure, not a primary data image. That distinction matters, because several publishers now treat generative images, and graphical abstracts in particular, under rules that are tighter than “add a caption and hope.” Draft the layout if it helps you think, then read the target journal before you treat the draft as a submission file.

A longer, tabulated version of the official language we were able to open is in the journal policy guide. A shorter narrative overview is in the journal policies article. This piece is about graphical abstracts only.

An explanatory summary, not a substitute for the data

A useful graphical abstract shows the story of the paper in one frame: the population or system, the intervention or exposure, the process, and the outcome you actually measured. It should be readable at a small size, with a handful of labels, and it should not introduce findings that are not in the text.

It is not a place to regenerate a blot, a micrograph, a radiology slice, or a photograph of a participant. Those are primary research images. Elsevier’s official policy, checked on 14 November 2026, states that AI tools must not be used to create or alter primary research images such as microscopy, histology, western blots, radiology scans, or patient images. Other publishers are at least as careful about not misrepresenting evidence. If a result can only be shown honestly as the original data figure, leave it in the results and do not “illustrate” it by generating a look-alike.

This restriction on illustrating or altering evidence is separate from AI used as a declared research method, such as AI-assisted imaging. Elsevier addresses that use separately and requires reproducible Methods documentation; it does not authorize a general image generator to invent experimental observations.

Journal rules are not interchangeable

Publisher policies differ, and they can change. The notes below reflect official pages opened on 14 November 2026. They are a map, not a substitute for the journal’s own instructions to authors.

Elsevier. Explanatory images may be created with AI tools if you disclose the specific tool, version, and how it was used in the caption and in a general AI declaration. Graphical abstracts are called out separately. Elsevier states that general-purpose generative AI image tools must not be used to create graphical abstracts, and it encourages dedicated scientific illustration tools. Its FAQ names BioRender and Mind the Graph as examples. Medical Illustration does not claim to be on that example list. If you are submitting a graphical abstract to an Elsevier journal, confirm what the journal will accept before you invest in a prompt-based draft.

Springer Nature. The current manuscript-preparation policy uses a risk-assessment framework. Visual content may be permitted when it comes from verifiable, attributable inputs, is reviewed by a person, represents the underlying evidence, and is disclosed. Opaque AI visuals created from text prompts without verifiable source material are not permitted. A prompt-only graphical abstract is likely to fall on the wrong side of that test unless you can show the editor how the image was derived from your data or from other verifiable inputs.

Nature research figure guide. The image-integrity page states: “The use of any sort of generative AI in figures is not permitted. This includes content-aware editing tools in Photoshop.” If you are submitting to a Nature journal, treat that wording as binding unless the journal tells you otherwise in writing.

Science journals (AAAS). The official editorial-policy page returned HTTP 403 during the 14 November 2026 review. Its current AI-image and disclosure rules are not verified here. Open the official page or ask the journal before preparing a submission; policies of Science Partner Journals do not establish the rules for all Science journals.

ICMJE recommendations. Journals that follow ICMJE should require disclosure of AI-assisted technologies, including image creators, at submission, in the cover letter and in the work. Figure generation belongs in the Methods when it applies. AI tools cannot be listed as authors. Humans must review the output.

Official links and the access date are collected on the journal policy guide. JAMA and Cell Press are outside this comparison.

Disclose the tool and the human work

Where a journal allows a graphical abstract that involved AI, expect to name the tool and version, describe how it was used, and state that a person reviewed and edited the result. Elsevier’s official page includes suggested caption language for an image that was generated and then edited by the authors; if you submit there, prefer the journal’s current wording to a paraphrase.

Keep a short record as you work: the prompt or source materials, the software version, the date, and the name of the person who checked the labels against the manuscript. If you later change the finding in the text, change the graphical abstract and review it again. Do not leave a decorative molecule or a mislabelled pathway in the abstract that the paper itself does not support.

A commercial-use license is a separate question. Paid Medical Illustration plans include one, as listed on pricing and the license page on 14 November 2026. That license does not override a publisher’s ban or a dedicated-tool rule.

Write the brief for the mechanism, not the decoration

If you are still at the drafting stage, or if your journal allows a reviewed explanatory summary, write a brief that could stand as a figure legend:

  • The study question in one clause.

  • The system (cell type, organ, cohort) and the intervention or comparison.

  • The process you want in the middle of the frame, in the order it occurs.

  • The outcome you measured, with no extra implied effect.

  • The view: a simple left-to-right pathway, a cycle, or a before/after pair, and the maximum number of labels.

  • What to leave out: decorative anatomy, invented receptors, and any structure not discussed in the paper.

Then generate a draft in the graphical abstract generator and check it against the text, not against a taste for colour. If a label names a pathway the paper does not study, remove it. If the layout is clearer as three panels than as one collage, split it.

Empty prompt words such as “accurate,” “realistic,” or “publication style” do not check the science. If the first draft is wrong, name the error in the next prompt.

A simple decision path

  1. Open the target journal’s current official policy and the graphical-abstract instructions.

  2. If generative figures are not permitted, or are permitted only with an editor’s prior agreement, stop or write to the editor before you generate a submission file.

  3. If you are submitting to an Elsevier journal, read the dedicated-tool rule for graphical abstracts and do not assume a prompt-based draft qualifies.

  4. If a reviewed explanatory summary is allowed, draft from the manuscript, review every label, and disclose the tool as required.

  5. Never generate or alter a primary research image to decorate the abstract.

Create a draft for review

Use a graphical-abstract draft to test the story of the paper. Decide whether you can submit it only after you have read the journal’s current official page.

Open the graphical abstract generator · Read the journal policy guide

Official sources and access notes