How to Disclose AI-Generated Images in Academic Papers

AI images make disclosure harder than text

Text disclosure already causes confusion. Images cause more.

A paragraph drafted with an AI tool still enters the paper as prose. A figure, diagram, graphical abstract, or synthetic image can look like evidence. Readers may not know whether they see measured data, a human drawing, a simulated scene, or a model output.

That difference matters.

If you used Midjourney, DALL-E, Stable Diffusion, Adobe Firefly, Canva AI, ChatGPT image generation, or another image tool, you should record the use before submission. Do not wait for the journal form. By then, you may have forgotten the prompt, tool version, editing steps, and reason for use.

An [[AI Usage Card](/chatgpt-disclosure-academic-papers/)](/what-are-ai-usage-cards/) gives you a simple place to record those details. You can generate one at ai-cards.org, then copy the text into your methods, acknowledgments, supplement, cover letter, or figure notes.

Start by naming the image type

Do not write one vague sentence that says "AI was used for figures." That sentence leaves editors guessing.

Name the kind of visual material.

A conceptual illustration differs from a data figure. A graphical abstract differs from a microscopy image. A cleaned plot differs from an AI-generated scene.

Use plain categories:

• Conceptual illustration
• Graphical abstract
• Schematic diagram
• Icon or decorative image
• Synthetic example image
• Edited or enhanced image
• Data visualization
• Research output from an AI model

This distinction helps because publishers treat these uses differently. Springer Nature tells authors to avoid generative AI images or figures for publication unless specific exceptions apply, such as legally sourced material, work directly about AI, or content based on verifiable scientific data that authors label as AI-generated. (group.springernature.com) Elsevier support material states that AI-generated or AI-altered figures, images, and artwork are generally not allowed, except in rare cases. (elsevier.support)

So the first question is not "Can I disclose this well enough?" The first question is: "Does this journal allow this kind of image at all?"

For the broader submission context, pair this guide with AI Transparency Requirements for Journal Submissions and [[[AI Disclosure](/ai-disclosure-in-systematic-reviews-and-meta-analyses/)](/how-to-disclose-ai-use-for-neurips-icml-and-acl-submissions/) Policies by Major Journals](/ai-disclosure-policies-by-journal/).

Separate illustration from evidence

Editors worry most when an AI image could pass as evidence.

A stylized picture of a cell, a classroom, or a legal archive may support communication. A fake gel, fake histology slide, fake satellite image, or fake clinical photograph can mislead readers if the paper presents it as observed material.

Keep this rule in mind: if the figure supports a claim about what happened in the world, readers need to know how you produced it.

JAMA guidance for medical publication allows AI use in research when authors describe it in the Methods section and take responsibility for all content, but it warns against presenting AI-generated data as if it came from non-AI sources. (jamanetwork.com) That principle travels well outside medicine. A figure that looks like data needs a clear production trail.

If your paper reports biomedical or clinical work, read the field guide on AI use in biomedical and clinical research papers before you submit.

Record the tool, prompt, date, and human edits

A useful image disclosure tells readers four things.

First, name the tool. Include the model or service if you know it. "Midjourney" says more than "AI image generator."

Second, state the purpose. Did you create a conceptual illustration? Did you draft a graphical abstract? Did you generate image variants for design exploration?

Third, record the prompt or describe it. You do not need to print a long prompt in the article body, but you should keep it in your records. A supplement can hold the prompt when it affects interpretation.

Fourth, describe human editing. If you changed labels, redrew parts, removed artifacts, adjusted colors, or combined the output with your own drawing, say so.

The Ecological Society of America gives a clear example of where disclosure can sit: authors who use AI tools in writing, image production, data collection, or analysis must disclose how and what tool they used in the manuscript section that matches the use, add a disclosure in the acknowledgments, and disclose it in the submission form. (esa.org)

That structure works for many journals even when the journal has not published detailed image rules.

Use figure captions for figure-level disclosure

If one figure uses AI and the rest do not, do not bury that fact in a general acknowledgment.

Put a short note in the figure caption. Then add a fuller statement in the methods, acknowledgments, or supplement.

A caption can say:

\begin{figure}
  \centering
  \includegraphics[width=0.85\linewidth]{conceptual-immune-pathway.pdf}
  \caption{Conceptual illustration of the proposed immune pathway.
  The authors generated an initial visual draft with Midjourney
  on 2026-08-14 using a text prompt describing the pathway structure.
  The authors redrew labels, adjusted layout, and verified that the
  figure does not present experimental data.}
  \label{fig:immune-pathway}
\end{figure}

That caption does useful work. It names the tool, date, use, edits, and evidentiary status.

If your journal wants a separate declaration instead, use the caption for clarity and the declaration for compliance. For LaTeX workflows, see the LaTeX Tutorial for AI Usage Cards and the Overleaf guide.

Put methodological AI image use in methods

Some image-related AI use belongs in methods, not acknowledgments.

If you used a segmentation model, image classifier, generative model, denoising tool, reconstruction tool, or AI-based enhancement step as part of the research process, describe it like any other method. Readers need enough information to assess the result and repeat the workflow.

A methods statement can look like this:

\subsection*{AI-assisted figure preparation}
 
The authors used DALL-E through ChatGPT on 2026-08-21 to create
three draft versions of a conceptual workflow diagram. The prompt asked
for a non-photographic schematic of the study design. The authors did
not use the generated image as data. The final figure was redrawn by
the authors in vector software, and all labels, arrows, and panel
relationships were checked against the study protocol.

For AI-generated or AI-assisted data figures, go further. Name the input data, software, model, parameters, validation checks, and any excluded outputs.

If AI touched the analysis pipeline, your disclosure should connect to your reproducibility materials. The disclosure log template in AI tools for research can help you keep that record during the project instead of reconstructing it after acceptance.

Add an AI Usage Card as a stable record

A journal statement often has to stay short. An AI Usage Card can hold the extra detail.

Use the card to record:

• Tool name and provider
• Date of use
• Version or model name, if available
• Prompt summary or full prompt
• Input materials
• Output used in the paper
• Human review and edits
• Location in the manuscript
• Whether the figure represents data, illustration, or communication

You can create the card at ai-cards.org. For example, if you used Midjourney to draft a conceptual figure for a humanities article, you could generate a card and then cite it in the acknowledgments:

\section*{Acknowledgments}
 
The authors used Midjourney to generate initial visual concepts for
Figure 2. The final figure was redrawn and edited by the authors.
No AI-generated image was used as evidence or source material.
An AI Usage Card documenting the tool, prompt summary, date, and
human edits is included in the supplementary materials.

If you need wording examples, see AI Usage Cards Examples and Templates.

Do not list the image tool as an author

An AI image tool cannot approve the final manuscript, answer questions about integrity, or take responsibility for errors.

COPE states that AI tools cannot be listed as authors because they cannot take responsibility for submitted work. (publicationethics.org) This applies to text tools and image tools alike.

You can credit the tool in a disclosure statement. Do not put "Midjourney," "DALL-E," or "Stable Diffusion" in the author list.

For more on authorship, see Can AI Be a Co-Author on a Research Paper?.

Check the tool terms, but do not confuse rights with disclosure

Tool terms and journal disclosure rules answer different questions.

OpenAI's terms say that, as between the user and OpenAI and to the extent allowed by law, the user owns input and output. (openai.com) Midjourney's terms, effective May 27, 2026, tell users that the service and generated assets come as-is and that users must decide whether use or redistribution is appropriate. (docs.midjourney.com)

Those terms do not decide whether a journal will accept an AI-generated figure. They also do not decide whether readers need disclosure.

Copyright adds another layer. The U.S. Copyright Office stated in 2025 that generative AI outputs can receive copyright protection only when a human author determines sufficient expressive elements, and that prompts alone do not qualify. (newsroom.loc.gov) If your project involves publication rights, permissions, licensing, or commercial reuse, ask your institution or a lawyer. For the manuscript, disclose the use either way.

Use this short decision test before submission

Ask these questions before you upload the manuscript:

Did the image tool create pixels, shapes, icons, layout, labels, or visual content that appears in the paper?

Could a reader mistake the image for observed data?

Did the AI tool alter an image that came from an experiment, field site, archive, patient, participant, or dataset?

Does the journal allow this type of AI image use?

Can you produce the prompt, date, output, and editing record if an editor asks?

If you answer yes to any of the first three questions, write a disclosure. If you cannot answer the last two, pause before submission.

If you still feel unsure, read Do I Need to Disclose AI Usage in My Paper?. The safest disclosure is specific, calm, and short.

A reusable disclosure template

Use this template and adjust it to your journal.

\section*{Declaration of AI-assisted image use}
 
The authors used [tool name, provider, version if known] on [date]
to [purpose, such as draft a conceptual illustration or generate
layout ideas for a graphical abstract]. The prompt requested
[brief prompt summary]. The AI output was used as [draft/reference/final
visual element]. The authors [describe human edits and verification].
The figure does not represent experimental, observational, clinical,
or participant data unless stated otherwise. The authors take
responsibility for the final figure and manuscript content.

For a figure note, shorten it:

AI use note: Initial visual draft generated with [tool] on [date].
Final labels, layout, and content were edited and verified by the authors.

Make the record before the editor asks

AI-generated images can save time during figure planning. They can also create trouble if readers cannot tell what they are seeing.

Treat every AI-assisted image as something you may need to explain.

Before you submit, generate an AI Usage Card at ai-cards.org. Use it as your working record, copy the relevant text into your manuscript, and attach the card as supplementary material when the journal allows it.

A clear card turns "we used AI for a figure" into a record that editors, reviewers, and readers can actually inspect.