03 · What You Need to Know
How AI-Generated Images Affect Research Disclosure and Image Integrity
What Counts as AI-Generated Research Imagery?
AI-generated research imagery includes visual material created or substantially modified through artificial intelligence systems and used in scholarly publications or research activities.
Examples include generated illustrations, conceptual artwork, graphical abstracts, synthetic photographs, scientific visualizations, and AI-modified experimental images.
However, not every image produced using software that incorporates AI should be treated identically.
For example, a researcher may use a generative model to create an imaginary classroom scene. Another may use an established image-analysis algorithm to segment cells in microscopy data.
The first creates new visual content. The second processes existing research data.
Both may require appropriate reporting, but their methodological roles and publisher restrictions can differ.
Why Are AI-Generated Images Treated Differently From AI-Generated Text?
Scientific images may function as direct evidence of observations, measurements, or experimental outcomes.
Readers may reasonably interpret a microscopy image, clinical photograph, or laboratory figure as representing something that was actually observed.
Generative AI systems can create convincing visual details without corresponding observations.
They may also introduce features during image enhancement that were absent from the original data.
Such changes can affect the evidentiary meaning of an image even when the result appears visually realistic.
Consequently, publisher policies often distinguish ordinary illustration from scientific image manipulation and data representation.
The relevant concern is not merely whether the image looks authentic, but whether its visual content accurately represents the underlying evidence.
What Does Springer Nature Say About AI-Generated Images?
Springer Nature's editorial policies restrict the use of generative AI-created images and artwork in publications, subject to specified exceptions and policy conditions.
Its guidance distinguishes generative AI artwork from certain AI-assisted image-processing activities and research uses in which AI image generation forms part of the scientific work.
Researchers should not assume that labeling an AI-generated illustration automatically makes it acceptable for publication.
Where a research exception applies, the journal may require details concerning the tool, procedure, and image generation process.
Authors should consult the current Springer Nature guidance and the specific journal's instructions before submitting generated artwork or figures.
The publisher's position illustrates why disclosure and permission must be assessed separately.
What Does Elsevier Say About Generative AI Images?
Elsevier maintains specific policies concerning generative AI and AI-assisted tools in journal publishing.
Its guidance restricts the use of generative AI to create or alter images in submitted manuscripts, with exceptions for circumstances in which AI use forms part of the research design or methodology and is appropriately described.
Publisher requirements may also distinguish scientific image processing from generative image creation.
Researchers should verify the current policy applicable to their intended journal, including requirements for graphical abstracts, artwork, and image manipulation.
It would be inaccurate to assume that a disclosure statement alone authorizes an image prohibited by the publisher.
How Do Major Publishing Positions Compare?
| Authority |
Relevant Position |
Practical Implication |
| Springer Nature |
Restricts generative AI-created images and artwork, with defined exceptions and conditions. |
Check whether the image is permitted before preparing a disclosure. |
| Elsevier |
Restricts generative AI image creation or alteration, with policy-specific research exceptions. |
Do not assume that labeling an image makes it acceptable. |
| ICMJE |
Requires appropriate reporting of AI-assisted technologies, including relevant figure-generation activities. |
Describe methodological AI image generation in methods as applicable. |
| COPE |
Emphasizes disclosure of AI use in image or graphical production and human responsibility. |
Report the tool's role according to the journal's instructions. |
These are general policy positions rather than a complete account of every journal's requirements.
Because image policies can change, authors should verify the latest instructions before preparing visual materials for submission.
What Is the Difference Between AI-Generated Artwork and Scientific Images?
AI-generated artwork typically serves an illustrative, explanatory, or aesthetic function.
A scientific image may represent an observation, measurement, experimental result, or analytical output.
Illustrative AI-generated imagery
Visual content created to communicate a concept or scenario without claiming to represent an actual research observation.
AI-generated or modified scientific evidence
Visual content presented as representing observations, measurements, or results that readers may use to evaluate research claims.
An AI-generated illustration of a hypothetical classroom does not ordinarily claim to document an actual participant or research event.
An AI-generated image presented as a photograph of participants during data collection would create a different and potentially serious integrity concern.
Even when the illustration is clearly conceptual, the publisher may still prohibit or restrict its use.
Can Researchers Use AI-Generated Illustrations in Academic Papers?
Possibly, but only where the publisher permits them and any required conditions are satisfied.
For example, a researcher may wish to illustrate the concept of human-AI collaboration using a generated image.
Such an illustration might be clearly labeled as conceptual rather than evidentiary.
However, some publishers restrict generative AI artwork regardless of whether it is intended as scientific evidence.
Researchers should therefore establish permission before generating or submitting the illustration.
Where permitted, the caption or designated disclosure should accurately describe its artificial origin and purpose.
Do AI-Generated Graphical Abstracts Require Disclosure?
Graphical abstracts may be subject to publisher policies governing both scientific figures and illustrative artwork.
An AI-generated graphical abstract can introduce inaccurate visual relationships, unsupported mechanisms, or misleading representations of results.
For example, a generated diagram might imply that an intervention caused improved learning outcomes when the study established only an association.
Researchers must verify the scientific meaning of every visual element.
They should also confirm that generative AI artwork is permitted for graphical abstracts under the journal's policy.
Disclosure does not substitute for either permission or scientific accuracy.
Does AI-Assisted Figure Design Require Disclosure?
The answer depends on what the AI tool did.
Using AI to suggest a color arrangement, generate plotting code, or recommend a clearer layout may be treated differently from generating new visual content.
For example, a researcher may ask ChatGPT to generate Python code for a bar chart using verified numerical results.
The resulting chart should be evaluated for accurate scales, labels, values, and data transformations.
The code-generation assistance may require disclosure under the journal's policy, particularly if it materially contributed to the research workflow.
However, producing a chart from genuine data using computational software is not automatically equivalent to generating an imaginary scientific image.
The distinction between AI-generated research code and AI-generated artwork is therefore important.
Can AI Be Used to Enhance Microscopy or Medical Images?
AI-assisted enhancement may be permissible in certain research contexts, but it requires careful methodological justification and compliance with image-integrity policies.
Some image-processing algorithms can reduce noise, improve resolution, or identify structures in existing data.
However, generative enhancement may introduce visual details that were not present in the original observations.
For example, an AI model might reconstruct apparent cellular structures in a low-resolution image even though the underlying data do not support those details.
Researchers should distinguish transformations that preserve or analyze recorded evidence from processes that synthesize new content.
Where image processing is permitted, the methods should explain the procedure, and original data should be preserved according to applicable requirements.
Generated visual details should not be presented as directly observed evidence without appropriate qualification and methodological justification.
What Is the Difference Between AI Image Enhancement and Image Fabrication?
Image enhancement involves processing existing visual data to improve visibility or support analysis.
Image fabrication involves creating or presenting information as research evidence when it does not correspond to the observations or procedures claimed.
The distinction depends on the nature and representation of the transformation.
For example, applying a documented, permitted image-processing procedure is different from generating a realistic laboratory image and presenting it as an actual experimental result.
Nevertheless, an enhancement procedure can become misleading if it introduces features or selectively alters information relevant to the findings.
Researchers should preserve original image data and report consequential transformations accurately.
The presence of AI does not automatically establish fabrication, but it can create risks that require specific safeguards.
Can AI-Generated Images Be Used as Research Stimuli?
Yes, where the research design, ethical requirements, and publication policies permit their use.
For example, an educational technology researcher might generate hypothetical classroom scenes to examine how students perceive AI-created instructional materials.
In this situation, the generated images are part of the experimental materials rather than photographs of actual events.
The methods section should describe their generation, selection, presentation, and relevant characteristics sufficiently for readers to understand the study.
Researchers may also need to consider whether the generated stimuli contain unintended stereotypes, unrealistic features, or other characteristics that could influence participants' responses.
The use of AI-generated stimuli should be distinguished clearly from presenting synthetic imagery as authentic research observations.
What About AI-Generated Synthetic Images Used for Model Training?
Synthetic images may be used in machine-learning research to augment datasets, test algorithms, or investigate model performance.
Such uses can be legitimate research methods when appropriately designed and reported.
However, synthetic data may not reproduce the distribution, variability, or characteristics of real-world observations adequately.
Researchers should explain how the images were generated, how they were used, and how the resulting models were evaluated.
For example, a model trained partly on synthetic images should be assessed using suitable evaluation data that support the intended claims.
The publication should not imply that synthetic observations were collected from real participants or physical experiments.
Applicable journal policies may provide exceptions for AI-generated images that are integral to the research methodology.
Can AI-Generated Images Be Presented as Photographs of Research Participants?
Not as authentic photographs of actual participants when no such photographs were taken.
Presenting synthetic imagery as documentation of real research events would misrepresent the evidence.
Researchers may sometimes use clearly labeled illustrations or synthetic examples for educational or methodological purposes, subject to the publisher's policy.
However, those images must not be described in a manner that leads readers to believe they document actual participants, observations, or procedures.
Where privacy considerations prevent publication of genuine participant photographs, generating fictional replacements does not automatically justify presenting them as original research data.
The distinction between illustration and evidence should remain explicit.
Does AI Image Generation Raise Copyright or Licensing Concerns?
Potentially. Copyright protection and the legal treatment of AI-generated material depend on the applicable jurisdiction and the nature of human creative contribution.
For example, the US Copyright Office has explained that purely AI-generated material may not qualify for copyright protection without sufficient human authorship.
Other jurisdictions may apply different legal frameworks.
Researchers should also consider the terms of the image-generation service and any publisher requirements concerning rights and permissions.
Permission to use a generated image under a service agreement does not necessarily establish that the intended journal permits it.
Copyright, licensing, image integrity, and AI disclosure are related but distinct questions.
What If AI Generates an Image Resembling a Real Person?
Generated images depicting identifiable individuals may raise concerns involving privacy, consent, likeness, and misleading representation.
For example, a synthetic image resembling a real student or faculty member could imply participation in research or endorsement of a publication.
Researchers should avoid creating misleading representations of identifiable individuals and comply with applicable consent and legal requirements.
Even fictional images may require appropriate labeling when their realism could cause readers to mistake them for actual research photographs.
Where a generated image is used as a research stimulus, the study should account for any relevant ethical implications of the depicted content.
Can AI-Generated Images Contain Bias or Stereotypes?
Yes. Generative image systems may reproduce stereotypical representations or unevenly depict demographic groups, professional roles, and social settings.
For example, a generated image of a scientist may repeatedly depict a narrow demographic profile, even when the research concerns a diverse population.
Such representations can affect the appropriateness of illustrations and, in experimental studies, may influence participant responses.
Researchers should evaluate generated images for unintended characteristics relevant to the research question.
Where synthetic images form part of experimental materials, these characteristics may need to be controlled or reported.
The visual plausibility of an image does not establish its representational appropriateness.
What Information Should an AI Image Disclosure Include?
Where image generation or modification is permitted and reportable, the disclosure should identify the tool's role and provide information appropriate to the image's function.
| Reporting Element |
Purpose |
| AI tool or model |
Identifies the system used to generate or modify the image. |
| Image purpose |
Distinguishes illustration, research stimulus, analytical output, or scientific evidence. |
| Generation or processing procedure |
Explains how the image was produced or transformed. |
| Original data source |
Identifies the underlying observations where image processing involved research data. |
| Human review |
Describes how accuracy, appropriateness, or integrity was evaluated. |
| Labeling and limitations |
Prevents synthetic or altered material from being mistaken for authentic observations. |
Additional details, including prompts, model versions, settings, and original images, may be required for particular research designs or publisher policies.
The statement should accurately describe the procedure rather than simply identify the image as AI-generated.
Where Should AI-Generated Images Be Disclosed?
The reporting location depends on the image's role and the journal's instructions.
AI-generated research stimuli or images produced as part of a computational method generally require methodological reporting.
Permitted illustrative artwork may require labeling in the figure caption or a designated AI-use statement.
AI-assisted processing of experimental images may require a detailed description in methods, together with preservation of the original data.
Some journals also require declarations during submission.
Researchers should determine the permitted use before deciding where to place the disclosure.
The broader question of where AI use belongs in a manuscript should be considered alongside the journal's image-integrity policy.
What Should an AI-Generated Image Disclosure Say?
Illustrative Methods Disclosure
AI-Generated Images Used as Research Stimuli
"Synthetic classroom images were generated using an AI image-generation system for use as experimental stimuli. The research team reviewed the images for consistency with the intended experimental conditions and excluded outputs containing unintended visual features. The images were presented to participants as synthetic materials and were not represented as photographs of actual classroom environments."
This statement is appropriate only if the described procedures actually occurred and the journal permits the use.
The tool, model version, generation procedure, and additional information should be reported where required.
Researchers should not use a generic declaration to conceal material differences among image generation, enhancement, and manipulation.
Does Labeling an Image as AI-Generated Make It Acceptable?
No. Disclosure does not override a publisher's restrictions.
A journal may prohibit generative AI artwork even when the image is clearly labeled.
Similarly, an image representing fabricated experimental observations cannot be made legitimate simply by adding a disclosure after the fact.
Researchers must establish that the image is permitted, that it accurately represents its intended function, and that the reporting requirements are satisfied.
Transparency is important, but it does not substitute for compliance with image-integrity standards.
Should Researchers Preserve Original Images and Generation Records?
Where AI processes scientific images, preserving original unaltered data is particularly important for evaluating the integrity of the resulting figures.
Researchers may also need to retain processing scripts, software settings, model information, and records of transformations.
For generated research stimuli, relevant records may include prompts, selected outputs, excluded alternatives, and documentation of experimental presentation.
The appropriate records depend on the research design and applicable institutional requirements.
Not every conceptual illustration requires the same documentation as an experimental image used to support scientific conclusions.
However, the inability to distinguish original observations from generated content can create serious problems for research verification.
What If the Journal's AI Image Policy Is Unclear?
Researchers should review the journal's image-integrity instructions, publisher-level AI guidance, and submission requirements.
Where uncertainty remains, the editorial office should be consulted before submitting consequential AI-generated or AI-modified imagery.
The inquiry should identify the image's purpose and whether it was generated from prompts, modified from existing research data, or produced through a documented analytical procedure.
These distinctions are more informative than asking whether the journal generally permits AI images.
Researchers should retain any clarification received and ensure that the final image accurately represents its intended role.
