03 · What You Need to Know
When AI-Generated Images Are Permitted in Scholarly Publications
Why Does the Type of Image Matter?
Research papers contain images with different purposes. Some explain concepts, others visualize analyzed data, and some document observations obtained during experiments or fieldwork.
These differences are consequential because a conceptual illustration does not carry the same evidential claim as a microscopy image, medical scan, or photograph of a research setting.
For example, an AI-generated diagram illustrating a proposed educational technology framework is a constructed representation. A photograph presented as documenting an actual classroom is an empirical record.
Generating the first may be appropriate under certain publication policies. Generating the second and presenting it as an authentic observation would misrepresent the research.
The broader question of creating images with generative AI for research should therefore be distinguished from the narrower question of whether those images are acceptable in a submitted manuscript.
What Does Elsevier's Current Policy Allow?
Elsevier's updated generative AI policy for journals distinguishes explanatory images, research and data images, and graphical abstracts or cover artwork.
Under its policy updated in 2026, authors may use AI tools to support certain explanatory images, including flowcharts, conceptual diagrams, decision trees, timelines, and experimental workflow illustrations. Such use requires human verification and appropriate disclosure.
Data visualizations may use AI assistance when they are directly derived from the underlying research data through reproducible analytical or computational methods. AI must not invent data or generate visual results unsupported by the research.
Primary research images are subject to stricter restrictions. The policy prohibits using AI to create or alter images representing primary observed or experimental data that were not directly obtained through the research.
Elsevier also distinguishes graphical abstracts from ordinary explanatory figures. Its journal policy prohibits general-purpose generative AI image tools for creating graphical abstracts and directs authors toward dedicated scientific illustration tools.
| Image Category |
Elsevier Journal Policy |
Key Requirement |
| Explanatory illustrations |
Permitted in specified circumstances |
Verify accuracy and disclose AI use in the caption and general declaration. |
| Data visualizations |
Permitted with conditions |
Derive figures directly from data using reproducible methods and document relevant AI assistance. |
| Primary scientific images |
Generative creation or alteration of primary observed data is prohibited |
Preserve authentic observations and follow applicable scientific image-processing requirements. |
| Graphical abstracts |
General-purpose generative AI image tools prohibited |
Use permitted scientific illustration approaches and verify the journal's requirements. |
| Journal cover artwork |
May be permitted with prior approval |
Obtain permission from the editor and publisher before submission. |
This summary concerns Elsevier's journal policy, not a universal publishing standard. Elsevier maintains separate policies for books and commissioned content, which may be more restrictive. Individual journal instructions should also be checked.
What About Springer Nature's Policy on AI-Generated Images?
Springer Nature maintains editorial guidance addressing generative AI images and artwork. Its approach has included restrictions on generative AI images in publications, with specific treatment of research methods and other defined circumstances.
Researchers should not assume that a generated illustration permitted by another publisher will automatically be acceptable in a Springer Nature journal.
The relevant policy should be checked directly, including any distinction between AI-generated artwork and images produced through an established research methodology.
Where a manuscript includes a generated visual, authors should establish whether the intended use falls within an expressly permitted category rather than relying on general statements that AI assistance is allowed.
Can AI-Generated Conceptual Diagrams Be Published?
Potentially. Conceptual diagrams are among the more defensible applications because they represent relationships, processes, or theoretical constructs rather than claiming to document observations.
For example, an educational technology researcher might create a diagram showing the proposed relationships among perceived usefulness, self-regulated learning, and intention to continue using a platform.
However, the diagram must accurately represent the researcher's framework. AI may introduce unsupported arrows, imply causal relationships, or mislabel constructs.
Researchers should verify every component and relationship before submission.
For further methodological considerations, creating conceptual research diagrams with AI involves checking whether the visual faithfully represents the theoretical and analytical structure.
Can AI-Generated Images Be Used as Scientific Figures?
The term "scientific figure" covers several distinct categories, so the answer depends on what the figure represents.
A schematic diagram illustrating a proposed mechanism is different from a figure showing measured experimental results.
Researchers may use AI assistance to construct certain explanatory figures where permitted. However, a generated image should not be substituted for an actual microscopic observation, laboratory photograph, radiological scan, or other empirical image.
Even when a synthetic scientific-looking image is visually convincing, it does not establish that the depicted structures or events were observed.
Researchers should also distinguish generating a figure from enhancing an existing scientific image, since processing authentic observations introduces additional integrity requirements.
Can AI Be Used to Generate Graphs and Charts for Publication?
Yes, in circumstances where the journal permits AI assistance and the visualizations are faithfully derived from actual research data.
For example, AI may help generate code that produces a bar chart from a verified dataset. The researcher can then execute the code, check the plotted values, and document the analytical procedure.
This differs from asking a general-purpose image generator to create a graph based on a textual description and accepting the resulting bars, labels, and values without verification.
Generated graph images may contain invented values, incorrect axes, or misleading visual relationships.
The separate task of creating research figures and graphs with generative AI requires a reproducible connection between the underlying data and the final visualization.
Can AI-Generated Images Be Used in a Graphical Abstract?
Not necessarily. Graphical abstracts are subject to publisher-specific rules that may differ from those governing figures within the manuscript.
Elsevier's updated journal policy, for example, prohibits general-purpose generative AI image tools for graphical abstracts and recommends dedicated scientific illustration tools.
Other publishers and journals may apply different requirements.
Researchers should check whether AI-generated elements are permitted, whether particular illustration tools are accepted, and what disclosure or licensing conditions apply.
These distinctions become important when considering AI-generated graphical abstracts for journal submission.
What if the AI Image Is Part of the Research Method?
An image generated as part of a research method may be treated differently from an image generated merely to decorate a manuscript.
For example, a study might investigate how participants respond to AI-generated faces, or evaluate a computer vision model using synthetic images.
In such cases, the generated images may be experimental stimuli or research objects. Their generation is part of the methodology and should be described accordingly.
Researchers should report the model or tool used, relevant generation procedures, selection criteria, and validation steps to the extent necessary for reproducibility and methodological assessment.
However, a methodological justification does not automatically override all journal policies. The intended publication venue may still impose requirements concerning how generated images are reproduced in the article.
Does Labeling an Image as AI-Generated Make It Acceptable?
No. Disclosure is necessary in many publication contexts, but it is not sufficient to establish acceptability.
An AI-generated microscopy image cannot become authentic experimental evidence simply because its caption identifies the generation tool.
Likewise, a scientifically inaccurate illustration does not become appropriate merely through transparency about its provenance.
Researchers must satisfy the relevant standards for the image's intended purpose, including accuracy, rights, ethical suitability, and editorial policy.
How Should AI-Generated Images Be Disclosed?
Disclosure requirements vary among publishers, but Elsevier's updated journal policy provides a concrete example.
For permitted explanatory images, it requires disclosure in the figure caption and the manuscript's general AI declaration. The caption should identify the tool and relevant details about how it was used.
For AI-assisted data visualizations or images generated as part of the research methodology, the Methods section should describe the procedure in a reproducible manner.
The International Committee of Medical Journal Editors also recommends transparency about AI-assisted technologies, including image creators, and emphasizes human author accountability.
Researchers should not assume that a generic acknowledgment such as "AI was used" provides sufficient information.
Illustrative Disclosure
Caption for a Permitted AI-Assisted Conceptual Figure
"Figure 1. Conceptual representation of the proposed educational technology framework. The initial illustration was generated using [tool and version] and subsequently revised by the authors. All constructs and relationships were checked against the study's theoretical framework."
This is an illustrative statement, not a universal publisher-approved disclosure template. Authors should adapt it to the actual process and the target journal's requirements.
The broader question of disclosing AI-assisted image processing may require additional information about the original images and the transformations applied to them.
What Copyright and Licensing Issues Should Researchers Consider?
AI-generated images raise questions about copyright ownership, platform terms, third-party rights, and permissions for publication.
The legal treatment of generated content varies by jurisdiction and the extent of human creative contribution.
In the United States, the Copyright Office has explained that purely AI-generated material generally lacks the human authorship necessary for copyright protection, although qualifying human-created elements may be protected.
Researchers should not assume that permission to use an AI platform's output means they own copyright in the image or that the image is free of third-party rights.
Some publishers may require authors to confirm that they possess the rights needed to reproduce every figure.
These questions should be resolved before submission, particularly when generated images resemble identifiable people, existing artworks, logos, or proprietary materials.
What if an AI-Generated Image Resembles Real Research Data?
This presents a serious research integrity concern when the image is represented as an authentic observation.
For example, a generated image resembling a Western blot or microscopic specimen should not be presented as a laboratory result obtained during the study.
Similarly, an AI-generated classroom photograph should not be used as documentary evidence of actual fieldwork.
Research integrity standards distinguish legitimate illustration from fabrication or falsification of research records.
Where AI is used to modify empirical images, researchers should examine whether the processing alters or invents scientifically meaningful information.
The risk of AI image enhancement crossing into research falsification is especially consequential when generated content is presented as observed data.
Watch Out
Do not assume that a publisher's permission to use AI for manuscript writing also permits AI-generated figures. Image policies may impose separate restrictions, and the rules for explanatory illustrations, research data images, and graphical abstracts may differ substantially.
Should Researchers Preserve the Original Images and Generation Records?
Yes, where relevant to the research and consistent with applicable data-retention requirements.
For AI-assisted figures derived from empirical data, original datasets, images, analytical scripts, and processing records should be retained so the final visual can be verified.
For generated illustrations or experimental stimuli, retaining the model information, prompts, generation settings where available, selected outputs, and substantive editing decisions can support transparency.
Researchers should also consider whether the generation process is reproducible. Identical prompts may not produce identical images, especially when model versions, settings, or random generation processes differ.
The separate question of preserving original scientific images is particularly important when the final figure derives from authentic observations.
04 · A Practical Example
Deciding Whether an AI-Generated Figure Can Be Submitted to a Journal
Hypothetical Example
A Conceptual Illustration for an Educational Technology Study
A researcher develops a conceptual framework examining how perceived usefulness, self-regulated learning, and digital learning experiences relate to students' intention to continue using an educational platform.
The researcher uses a generative AI tool to create a visual illustration of the framework and plans to include it as Figure 1 in a journal manuscript.
The target journal follows Elsevier's updated generative AI policy for journals.
Step 1: Classify the Figure
The researcher determines that the image is a conceptual illustration rather than an empirical photograph or data visualization.
It represents proposed theoretical relationships and does not claim to document observed participants or measured outcomes.
Step 2: Check the Applicable Publisher Policy
Under Elsevier's updated journal policy, certain explanatory images may be created with AI assistance when the relevant conditions are satisfied.
The researcher confirms the current journal instructions and whether any additional requirements apply to conceptual figures.
Step 3: Verify the Scientific Content
The AI-generated illustration contains arrows implying causal relationships between several constructs.
However, the researcher's theoretical framework proposes associations rather than established causal effects.
The researcher removes or revises the misleading arrows and checks every construct label against the actual framework.
Step 4: Prepare the Disclosure
The researcher documents the generation tool, version, purpose, and subsequent revisions.
A disclosure is included in the figure caption and the manuscript's AI declaration according to the applicable journal requirements.
Step 5: Retain the Supporting Records
The researcher preserves the original framework, generated image, revised figure, and documentation of the changes.
The final figure can then be evaluated as a permitted explanatory illustration, provided the journal's requirements are satisfied.
The key distinction is that the image communicates the researcher's conceptual framework rather than inventing or replacing empirical evidence.
