Can You Use AI-Generated Figures in a Journal Article?

Publishers don't rule on "AI images" as one thing. They sort figures by what the figure claims to show, and the rules follow from that. An explanatory figure shows concepts, processes or relationships rather than measured data; that is the kind of figure Elsevier, Wiley and Taylor & Francis let AI help with. A data visualization is a chart or plot computed from your measurements. A primary research image is a direct record of what you observed, such as a micrograph or a blot, so it counts as evidence.

The safest AI figure is an explanatory one built from content you can show: your variables and hypotheses, your screening counts, your methods.

Quotes are from each publisher's own page, checked on 19 September 2026. Individual journals can be stricter than their publisher — Taylor & Francis notes that some of its journals allow AI only for language editing — so read your journal's author instructions before you submit.

Science adds that "exceptions may be granted in certain situations—e.g., for images and/or videos in manuscripts specifically about AI and/or machine learning." Its editorial policies page points to separate AI usage guidelines for where to declare each kind of use.

Older articles, including a June 2025 Nature Index news story on nature.com, say Nature does not permit AI-generated images; Nature Index news is editorially independent of Springer Nature, so it is not a policy statement. Nature Portfolio's current AI policy page no longer states a blanket ban: it uses Springer Nature's green, amber and red risk tiers (checked 19 September 2026). Under those tiers, figures derived from your own content can be allowed, but "creating content from text prompts" is still not permitted — see the Springer Nature row.

Elsevier, Springer Nature and Taylor & Francis want the disclosure in two places. Wiley's figure guidance asks for the caption; MDPI, Frontiers, IEEE and PLOS ask for it in the article text, and ACM asks for a Methods description when the figure is relevant to your conclusions; Science wants it in your cover letter and in the Methods or Acknowledgments. The two places are:

1. The figure caption or legend — which tool made the figure, which version or date, and what it did. Elsevier, Springer Nature, Wiley and Taylor & Francis ask for this.
2. A statement in the article — an AI declaration section, the Acknowledgments or the Methods, depending on the publisher.

For Vizcept, the "version" a reviewer can check is the underlying model and the month you used it. Image-model figures are drawn by Google's Nano Banana Pro; revisions you make in the Vizcept workbench are drawn by Google's Nano Banana 2. Syntax trees and sentence diagrams are normally analysed by a language model (Anthropic's Claude Sonnet 4.6, with OpenAI's GPT-5.2 as backup) and typeset by Vizcept's own layout code (how that works). A tree made from a sketch, from text with characters the font can't set, or from an analysis that fails is drawn by the image model instead, and the result doesn't yet show which way it went.

Pick your publisher below to get both parts with the details its policy asks for. The output is in English because the journals publish in English. Edit it to match your figure number and what you actually did.

Publishers' own examples, quoted as published (checked 19 September 2026).

Elsevier: AI declaration. It goes in a section titled "Declaration of generative AI and AI-assisted technologies in the manuscript preparation process":

During the preparation of this work, the author(s) used [NAME OF TOOL / SERVICE] in order to [REASON]. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the published article.

Elsevier: figure caption.

Figure 1. Schematic overview of study design. Initial image generated using [NAME OF TOOL/SERVICE] and subsequently edited by the authors. All elements were reviewed for accuracy and completeness.

Wiley: figure caption.

Figure 1. Conceptual diagram of treatment pathway. Diagram created using DALL·E 3 (accessed March 2024) to generate an initial draft, with subsequent modifications by the authors in Canva. All content was reviewed for accuracy by the clinical team.

MDPI: statement.

During the preparation of this manuscript/study, the author(s) used [tool name, version information] for the purposes of [description of use]. The authors have reviewed and edited the output and take full responsibility for the content of this publication.

Sources: Elsevier policy, Wiley AI guidelines, MDPI ethics policy. The generator above fills these in for a Vizcept figure.

It depends on your university, not the publisher. Graduate schools set their own AI rules, and some departments are stricter than their university. Before you put an AI-assisted figure in a thesis:

• Ask your supervisor, and check your graduate school's generative AI guidance.
• Draw only explanatory figures this way — never figures that stand for your data.
• Add a figure note saying which tool you used, when, and that you checked it, and mention it wherever your university wants AI use declared.
• Keep what the figure was built from (your variable list, your screening log) with your thesis files.

The disclosure statement generator above has a thesis option that writes a figure note in plain academic style.

Images that stand for evidence. That means western blots, gels, micrographs, histology, radiology and patient images, and any "result" picture that should come from your own instruments. No publisher above that addresses these images accepts AI-generated versions.

Edits differ by publisher. Elsevier bars AI tools from altering these images at all, even for brightness, contrast or color balance, and Wiley says AI "must not be used to generate, modify, or enhance" factual and evidential images. Taylor & Francis bans generative AI from manipulating images; other changes must be minor and applied to the whole image. Springer Nature says non-generative machine-learning tools that "analyse, improve, combine, restore, segment or enhance existing visual content may be permitted" if the material stays attributable, nothing is misrepresented and the use is disclosed.

Plots of your own data belong in the software that computed them — R, Python, Prism or similar — so the figure can be reproduced from the data. Vizcept is for explanatory figures, and its terms forbid using it to create images of experimental data.

• Conceptual frameworks, PRISMA flow diagrams and workflow diagrams are explanatory figures, built from content you supply (your variables, your counts) and drawn by an image model from your description. Elsevier, Wiley and Taylor & Francis allow this with disclosure. Springer Nature allows "diagrams derived from author-developed content" but not "creating content from text prompts", so for Springer Nature and Nature Portfolio journals, ask the editor first; Science journals need the editors' explicit permission. Proofread every label either way, and check every number in a PRISMA diagram against your screening log.
• Syntax trees and sentence diagrams are normally typeset by code from a language model's analysis of your sentence, so the words are set in a font instead of being drawn by an image model. If you start from a sketch, the analysis fails, or your sentence has characters the font can't set (such as Chinese, Japanese or Korean), Vizcept draws the figure with the image model instead. Check the analysis and every label.
• Graphical abstracts made with an AI image model should not go to Elsevier journals, and are worth checking with the editor anywhere else; the graphical abstract maker page explains where a draft still helps.
• Your rights: Elsevier and Wiley ask authors to check that a tool takes no training rights over their inputs and puts no limits on publishing the output; Taylor & Francis asks you to confirm you checked the tool's terms of use. Vizcept's publication rights page explains our terms.
• Comparing tools? If you mainly make a few figures a month, see our BioRender alternative comparison.