Can Generative AI Be Used to Create Images for Research?

Can Researchers Use Generative AI to Create Images Without Compromising Research Integrity?

You need an image for a research project. Perhaps you want to illustrate a theoretical concept, develop visual stimuli for an experiment, or communicate a scientific process that would be difficult to explain using text alone. A generative AI tool can produce a convincing image from a short description.

But can that image legitimately become part of your research? Does it matter whether the image is a conceptual illustration, an experimental stimulus, or something that appears to show an actual scientific observation?

The distinction matters because generative AI can create visually plausible content without observing the phenomenon it depicts. An image may be useful for communicating an idea while remaining unsuitable as evidence that the depicted event, object, or result actually exists.

Yes, but the Purpose of the Image Determines Whether Its Use Is Appropriate

What Generative AI Images Can and Cannot Contribute to Research

What Is an AI-Generated Research Image?

An AI-generated image is visual content produced wholly or partly through a generative model. Depending on the system, it may be created from a text prompt, an existing image, a reference image, or other conditioning information.

Such models can produce realistic photographs, diagrams, illustrations, and visual representations of objects or processes. Their outputs may resemble images captured by cameras, microscopes, or scientific instruments even when no corresponding observation occurred.

This makes it important to distinguish how an image was produced from what the image is intended to represent.

An AI-generated illustration of a laboratory can communicate what a laboratory environment might look like. It cannot establish that a particular experiment occurred in that laboratory.

What Types of Research Images Can Generative AI Create?

The potential uses are broader than producing illustrations for journal articles.

These categories may overlap. A generated image can function as an illustration in one study and as an experimental stimulus in another. Its methodological status depends on how the researcher uses it.

What Is the Difference Between an AI-Generated Illustration and Scientific Evidence?

The most consequential distinction is whether the image represents an observation or a constructed depiction.

For example, an AI-generated image depicting the interior of a classroom may be appropriate for illustrating a hypothetical learning environment. It would be misleading to present that image as a photograph documenting the actual conditions of a classroom included in the study.

Likewise, a generated image resembling a microscopic specimen cannot be substituted for an image acquired during laboratory analysis.

The visual plausibility of an image does not establish its empirical provenance.

Can AI-Generated Images Be Used as Experimental Stimuli?

Yes. Researchers may use generated images as stimuli in studies of perception, cognition, human-computer interaction, educational experiences, or related phenomena.

For example, an educational technology researcher might investigate whether students respond differently to instructional materials featuring AI-generated characters and materials featuring conventional illustrations.

The generated images become part of the experimental materials rather than evidence that the depicted scenes occurred.

Researchers should consider whether the stimuli differ in unintended ways, such as visual complexity, emotional expression, cultural representation, perceived realism, or background detail.

Such characteristics may influence participants' responses and complicate interpretation.

Where the design requires comparable stimuli, researchers should pilot-test or otherwise validate relevant image properties. Merely using similar prompts does not guarantee that generated images are equivalent.

Can Researchers Use AI to Generate Synthetic Image Datasets?

Yes, although synthetic images introduce methodological considerations that differ from ordinary illustration.

Researchers may generate artificial images to investigate computer vision methods, examine model robustness, or supplement particular training procedures.

However, synthetic data are not automatically representative of the population or phenomena they are intended to model.

Generated images may reproduce biases in the training data, contain unrealistic features, or fail to represent rare but scientifically important variation.

If synthetic images are used to train or evaluate a model, researchers should document the generation method, selection criteria, and relationship between synthetic and authentic data.

Evaluation exclusively on generated images may provide an incomplete account of performance on real-world observations.

Can AI Generate Scientifically Accurate Illustrations?

Sometimes, but scientific accuracy cannot be assumed from visual quality.

A generative model may produce an anatomically plausible illustration containing incorrect structures, a laboratory scene with impossible equipment arrangements, or a technical diagram with misleading relationships.

These errors can be difficult to recognize because the image may appear professionally rendered.

Researchers should inspect scientifically meaningful details against authoritative references, validated models, or established domain knowledge.

For precise technical diagrams, constructing the visual from verified components may be more appropriate than relying on unconstrained image generation.

The goal is not merely an attractive image. It is a representation that communicates the intended scientific information without introducing false details.

Can Generated Images Be Used to Represent Research Participants?

AI-generated people may be useful in hypothetical scenarios or research materials, but researchers should avoid implying that these images depict actual participants unless that representation is accurate and appropriately authorized.

For example, a generated student character may be used in a fictional classroom scenario. It should not be presented as a photograph of an interview participant.

Researchers should also consider whether generated faces or identities reproduce stereotypes or introduce characteristics that affect the study's interpretation.

If images are generated from photographs of identifiable individuals, privacy, consent, and data-processing requirements become particularly important.

Removing a real person's identifying features through generative editing does not automatically establish that the resulting image is anonymous or that its use is ethically permissible.

What Is the Difference Between Generating an Image and Enhancing an Existing Scientific Image?

Creating a new image and modifying an empirical image are distinct activities.

Generative image creation produces a constructed visual representation. Image enhancement begins with an existing image and changes some aspect of its appearance or content.

For scientific images, the distinction is consequential because enhancement may alter features that researchers intend to measure or interpret.

Operations such as denoising, sharpening, and upscaling may affect scientifically meaningful information, especially when performed using generative models that infer or synthesize visual detail.

The separate question of using AI to enhance scientific images requires consideration of image provenance, processing validity, and preservation of original observations.

Can AI-Generated Images Be Used in Research Publications?

Potentially, but acceptance depends on the publisher, journal, image category, and intended use.

Some publishers distinguish between generative AI images, conventional scientific figures, and images produced through established analytical methods.

For example, Springer Nature publishes guidance addressing AI-generated images and the circumstances under which their use may be restricted or permitted. Other publishers maintain their own policies.

Researchers should verify the current requirements of the intended publication venue before creating or submitting images.

The question of including AI-generated images in a research paper therefore cannot be answered solely by determining whether the image is scientifically accurate.

What Should Researchers Disclose About AI-Generated Images?

Disclosure requirements vary, but transparent documentation should generally allow readers or reviewers to understand the image's provenance and intended role.

Relevant information may include the tool or model used, the purpose of generation, whether the image represents a hypothetical scene, and any substantial editing or selection procedures.

For experimental stimuli or synthetic datasets, additional methodological documentation may be necessary to support reproducibility and evaluation.

A figure caption may need to identify the image as AI-generated, but a caption alone may not satisfy every journal's disclosure requirements.

Researchers should distinguish disclosing AI image processing from documenting images generated entirely from prompts, since the relevant methods and provenance differ.

Who Owns the Copyright in AI-Generated Research Images?

Copyright treatment depends on jurisdiction, the nature of human creative contributions, and the circumstances under which the image was generated.

The United States Copyright Office has explained that copyright protection requires sufficient human authorship and that purely AI-generated material is not protected merely because a person supplied a prompt.

Other jurisdictions may apply different rules or legal interpretations.

Researchers should also distinguish copyright ownership from permission to use an image under a platform's terms. Commercial-use permission does not necessarily establish copyright ownership, originality, or freedom from third-party rights.

When images are intended for publication or distribution, relevant licensing and rights questions should be resolved before use.

Can an AI-Generated Image Constitute Research Fabrication?

Yes, if generated visual content is knowingly presented as authentic research data or observations that did not occur.

For example, generating a microscopic image and presenting it as a specimen obtained during an experiment would misrepresent the evidential record.

Research fabrication concerns making up data or results and recording or reporting them as genuine. Whether a particular incident constitutes misconduct depends on the applicable definitions, evidence, and institutional procedures.

An AI-generated illustration that is accurately identified as hypothetical is fundamentally different from a fabricated empirical image.

The risk becomes particularly serious when researchers use generative editing to fill missing scientific image details and subsequently present the synthesized content as observed information.

How Can Researchers Verify AI-Generated Images?

Verification should reflect the intended use.

For a conceptual illustration, examine whether the visual accurately represents the intended theory or process. For an experimental stimulus, assess whether relevant properties are controlled and whether unintended differences could influence participants.

For synthetic datasets, evaluate representativeness, potential bias, and performance on authentic data where appropriate.

Researchers should retain generation records and document substantive decisions affecting the final image.

If the image is a scientific figure containing measurements, labels, or graphical relationships, the numerical and conceptual information should be verified independently.

AI-generated visuals should not be assumed accurate merely because the output is aesthetically convincing.

Using AI-Generated Images as Visual Stimuli in Educational Technology Research

Step 1: Define the Role of the Images

The researcher specifies that the images will appear in instructional materials shown to participants.

They are not intended to document actual classroom conditions or represent empirical findings.

Step 2: Generate and Inspect the Stimuli

The researcher generates candidate images and examines their visual characteristics.

Some images contain unintended differences in facial expression, clothing, background complexity, and perceived age. These differences could influence participant responses independently of the intended visual-style manipulation.

Step 3: Revise and Validate the Materials

The researcher selects or revises images to reduce unintended variation and conducts an appropriate pilot assessment of relevant stimulus characteristics.

The generation procedure, selection criteria, and validation decisions are documented.

Step 4: Use the Images in the Study

Participants view the instructional materials according to the experimental design.

The researcher analyzes their responses using the planned procedures. The resulting participant data, rather than the generated images themselves, provide evidence about how the stimuli were perceived.

Step 5: Report the Image Generation Procedure

The research report identifies the materials as AI-generated and describes the generation and selection procedures sufficiently for readers to understand the experimental manipulation.

The researcher also follows applicable ethics, copyright, and publication requirements.

This example illustrates a legitimate research role for generated images: they are controlled materials used to investigate a question, not fabricated observations presented as results.

Common Misconceptions About AI-Generated Images in Research

How to Decide Whether an AI-Generated Image Belongs in Your Research

Begin by defining what the image is supposed to accomplish. Its intended role determines the relevant standards of accuracy, validation, documentation, and disclosure.

A Practical Prompt for Creating a Research Illustration

Even with careful instructions, the output requires inspection. Researchers should correct inaccuracies and retain sufficient documentation to explain how the final image was produced.

The practical principle is straightforward: use generative AI to construct representations where construction is appropriate, but do not allow generated visual content to masquerade as observed evidence.

Before Using an AI-Generated Image in Research, Verify These

Frequently Asked Questions About AI-Generated Images for Research

AI Can Create Research Images, but It Cannot Create Evidence That Was Never Observed

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.