How to Disclose Midjourney Use in Academic Papers

Midjourney creates a paper trail, even when the image looks simple

A Midjourney image can look like a harmless illustration.

A diagram for a lecture paper. A cover image for a conference poster. A speculative rendering for an architecture article. A graphical abstract that makes a method easier to grasp.

Editors may see something else: an AI generated visual element that needs a record.

That record should say what you used Midjourney for, where the image appears, what prompt or source material shaped it, and what the authors checked before submission. If you already use an [[AI Usage Card](/chatgpt-disclosure-academic-papers/)](/what-are-ai-usage-cards/), Midjourney belongs there too.

You can generate one at ai-cards.org before you submit. Treat it as a clean source note for your figures, captions, acknowledgments, and cover letter.

When Midjourney use needs disclosure

Disclose Midjourney when it contributes to the scholarly object that readers, reviewers, or editors see.

That includes figures, graphical abstracts, cover art, icons, conceptual illustrations, visual stimuli, presentation images later reused in an article, and edited versions of existing images.

You should also disclose Midjourney when you use it during research design in a way that affects the work. For example, if you use Midjourney to create visual stimuli for a psychology experiment, the tool belongs in the methods section, not only in the acknowledgments. The same logic applies when you use generated images in education research materials, HCI prototypes, or qualitative elicitation tasks.

If you only use Midjourney to brainstorm mood boards that never enter the research process or manuscript, your journal may not require disclosure. Still, keep a private record. Policies change, and reviewers ask questions.

For wider image guidance, read How to Disclose AI-Generated Images in Academic Papers. That article covers image cases that go beyond one tool.

Where to put the disclosure

Put the disclosure where the reader can connect it to the image.

For most papers, that means the figure caption and an AI disclosure statement near the acknowledgments, methods, or declarations section.

A caption can stay short:

Figure 2. Conceptual illustration of the proposed classroom feedback loop. The authors generated the base image with Midjourney and edited labels, colors, and layout manually.

The disclosure statement can carry the details:

The authors used Midjourney to generate the base illustration for Figure 2. The prompt described a generic classroom feedback loop and did not include participant data, unpublished findings, or identifiable people. The authors selected one output, edited it manually, added all labels themselves, and take responsibility for the final figure.

That last sentence matters. COPE states that AI tools cannot serve as authors, so the named authors remain responsible for the content they submit. (publicationethics.org)

If your manuscript already contains an AI statement for writing support, do not bury Midjourney inside a vague line such as "AI tools were used during preparation." Name the tool and the task.

What to record before you submit

Midjourney disclosure gets easier if you save details while you work.

Record the tool name, date of use, prompt, model version or visible settings, source images, whether you used image references, whether you edited the output, and where the result appears in the manuscript.

Midjourney documents parameters such as aspect ratio and version as prompt controls, and those settings can change the output. If those settings affect a figure, record them with the prompt. (docs.midjourney.com)

A minimal record can look like this:

\section*{AI tool use}
 
The authors used Midjourney on 2026-10-18 to generate a conceptual illustration for Figure~3. The prompt described a generic laboratory workflow and did not include participant data, patient data, unpublished results, or identifiable individuals. The authors selected one generated image, edited the colors and layout in Inkscape, and added all labels manually. The final figure serves as an illustration only and does not present empirical data.

If you use the AI Usage Cards Examples and Templates page, you can adapt the wording for figure captions, declarations, or supplementary materials.

Do not use Midjourney for empirical images

Midjourney can create convincing images. That strength makes it risky in research.

Do not use it to create or alter data-bearing images. Do not use it to "improve" microscopy, radiology, satellite, field site, forensic, or clinical images. Do not use it to fill missing visual evidence. Do not use it to make a figure look cleaner when that figure claims to show observed data.

Springer Nature classifies AI use that generates data or figures as empirical content, or creates content from text prompts without verifiable source material, as not permitted in its manuscript preparation policy. (springernature.com)

That rule captures the main editorial concern: readers must know whether an image documents something that happened or depicts an idea.

Midjourney suits conceptual illustrations better than evidence. If a figure represents measured results, build it from the data, code, or original observations. Then disclose any software used for plotting, editing, or formatting.

For journal submission planning, pair this guide with AI Transparency Requirements for Journal Submissions.

If the image shows people, slow down

Midjourney can create synthetic people. That does not make the image harmless.

If your article concerns health, race, disability, gender, education, migration, crime, or other sensitive topics, a generated person can carry stereotypes into your work. A reader may also mistake a synthetic person for a participant, patient, or real community member.

Say so in the caption if the image depicts a synthetic person:

Figure 1. Synthetic image generated with Midjourney for illustrative purposes. The image does not show a participant, patient, or real research site.

If the image could influence interpretation, explain why you used a generated image rather than a photograph, diagram, or licensed stock image. Editors may ask.

If your work includes human participants, your institutional review board or ethics committee may also care about generated stimuli. Keep the AI record with your study materials.

Privacy and prompt records

Do not paste sensitive data into Midjourney prompts.

That includes participant descriptions, patient data, unpublished field notes, confidential client material, identifiable locations, or manuscript text under review.

Midjourney says Stealth Mode keeps creations hidden from other users on the website, but it also says images and videos generated in shared Discord spaces can still be visible to users in those channels. Midjourney limits Stealth Mode to Pro and Mega plans. (docs.midjourney.com)

That makes privacy a research workflow issue, not a caption issue.

If you need a visual based on sensitive material, write an abstract prompt that removes identifying details. Better yet, create the visual yourself or work with a designer under your institution's data rules.

Your disclosure does not need to publish every prompt if the prompt contains sensitive information. In that case, describe the prompt category and state why you withheld the exact text. Keep the full record in a secure project file if your ethics approval allows it.

Copyright and rights are separate from disclosure

Midjourney's own documentation says users own the images and videos they create, with exceptions such as upscaling an image created by another user. (docs.midjourney.com)

That does not answer every publication question.

Copyright law, publisher contracts, journal figure rules, third party rights, trademark concerns, and research ethics can still matter. The U.S. Copyright Office has said that AI can assist a human creative process without barring copyright protection for the human-authored work, but mere prompts alone do not create copyrightable authorship. (copyright.gov)

So do not write, "No disclosure is needed because I own the image." Ownership and transparency answer different questions.

A journal asks: what did the authors do, what did the AI system do, and can readers trust the figure?

Your disclosure answers that.

Midjourney disclosure for methods sections

Use a methods disclosure when the image plays a role in the research process.

That includes visual stimuli, interview prompts, design probes, training materials, classroom materials, annotation examples, and experimental tasks.

A methods statement should describe the generation process with enough detail for review:

\subsection*{Stimulus generation}
 
We generated 24 synthetic street-scene images with Midjourney on 2026-09-12 for use as interview prompts. Prompts described generic urban scenes and did not include photographs, participant data, or real addresses. We excluded outputs that contained readable text, distorted faces, brand marks, or identifiable landmarks. Two authors reviewed the final set before data collection. The images served as elicitation materials and not as observational data.

This level of detail helps reviewers assess bias, reproducibility, and participant risk.

Researchers in social science and HCI often need this distinction because AI images can shape participant responses. For field specific examples, see AI Disclosure for HCI and CHI Submissions and AI Disclosure for Social Science Research.

Midjourney disclosure for figure captions

Use a caption disclosure when the figure appears in the manuscript but does not affect the research design.

A caption should answer two questions: did AI help create this image, and what does the image represent?

\caption{Conceptual illustration of the proposed annotation workflow. The authors generated the base image with Midjourney and edited the layout and labels manually. The image is illustrative and does not present study data.}

For a graphical abstract:

\caption{Graphical abstract. The authors used Midjourney to generate the background illustration and assembled the final layout manually. All scientific claims in the graphic derive from the manuscript text and reported analyses.}

For a cover image:

\section*{AI image disclosure}
 
The submitted cover image was generated with Midjourney from an author-written prompt and then edited manually. It does not contain empirical data, participant likenesses, patient information, or unpublished third party material.

If the publisher gives exact wording, use that wording. Some submission systems ask for AI tool use in a separate field, and some journals want a declaration inside the manuscript file.

A short checklist before submission

Before you submit, open every image file and ask a plain question: would a reviewer want to know that Midjourney touched this?

If yes, check that you have the tool name, date, task, prompt summary, source material, edits, location in manuscript, and author review.

Then check the journal policy. Wiley states that authors who use generative AI to develop any portion of a manuscript should describe that use in detail in the Methods or Acknowledgements section, and authors remain responsible for accuracy. (onlinelibrary.wiley.com)

Also check whether the publisher permits AI generated visual content at all. Some policies allow conceptual illustrations with disclosure. Others draw stricter lines for figures, graphical abstracts, and visual material.

If the rule seems unclear, ask the editorial office before submission. Keep the answer with your manuscript records.

Use an AI Usage Card for Midjourney

A good Midjourney disclosure does not need to sound defensive.

It needs to sound precise.

An AI Usage Card gives you a reusable record of what happened: tool, purpose, input, output, human review, and placement in the manuscript. You can copy text from the card into a figure caption, methods section, acknowledgment, cover letter, or supplementary file.

If you already track writing tools with AI tools for research: a disclosure log template for papers and theses, add Midjourney to the same log. If you work in LaTeX or Overleaf, the LaTeX Tutorial for AI Usage Cards shows how to place disclosure text cleanly in a manuscript.

Generate an AI Usage Card at ai-cards.org before you submit. Then your Midjourney disclosure will not depend on memory, guesswork, or a last-minute scramble through old prompts.