AI-Generated Images in Academic Papers: Current Policies and Best Practices

Explore current publisher policies on AI-generated images in academic papers, learn to distinguish between acceptable and unacceptable uses, and discover alternative approaches for creating compliant visual content that meets scholarly standards.

In the rapidly evolving landscape of academic publishing, artificial intelligence has emerged as both a powerful tool and a subject of significant debate. As AI image generation technologies like DALL-E, Midjourney, and Stable Diffusion become increasingly sophisticated, researchers are exploring their potential applications in academic work. However, this technological advancement has prompted publishers, institutions, and ethics committees to reconsider and adapt their policies regarding the use of AI-generated content in scholarly publications.

This blog post examines the current state of publisher policies on AI-generated visuals in academic papers, distinguishes between acceptable and unacceptable uses of these technologies, and offers alternative approaches for creating compliant visual content that meets scholarly standards.

Analysis of Publisher Policies on AI-Generated Visuals

Major Publisher Stances

Academic publishers have been developing and refining their policies on AI-generated content at different rates, leading to a somewhat fragmented landscape of guidelines. Here’s how some of the major publishers currently approach AI-generated images:

Nature Portfolio and Springer Nature

Nature has established one of the more comprehensive frameworks for AI-generated content. Their policy specifies that:

• AI-generated images must be clearly labeled as such in figure captions
• Authors must provide a detailed description of the AI tool used, including version
• The specific prompts or parameters used to generate the images should be documented in the methods section or supplementary materials
• Authors retain responsibility for the accuracy and integrity of all content, including AI-generated visuals

Importantly, Nature distinguishes between AI tools used for data visualization (generally permitted with proper documentation) and AI tools used to create hypothetical imagery or artistic renderings (subject to more scrutiny).

Elsevier

Elsevier’s approach focuses on transparency and accountability:

• Authors must disclose the use of AI tools in their methodology section
• Generated images cannot substitute for experimental results or data visualizations where traditional methods exist
• The provenance of all visual elements must be clearly documented
• Elsevier reserves the right to request the original prompts or parameters used to generate images

PLOS

PLOS has taken a more cautious approach:

• AI-generated images are generally discouraged for critical research components
• If used, they must be clearly identified, with exhaustive documentation of methods
• Authors must confirm that the images do not contain fabricated or misrepresented data
• The use of AI cannot circumvent ethical requirements regarding human or animal subjects

IEEE

The Institute of Electrical and Electronics Engineers has specific requirements:

• AI-generated images must be clearly identified
• The AI system used must be credited appropriately
• Authors must verify that the images do not violate any copyright or intellectual property rights
• Images generated from proprietary datasets must include appropriate attributions

Policy Commonalities and Trends

Despite variations in specific requirements, several common threads emerge across publisher policies:

1. Transparency requirement: Nearly all publishers now require explicit disclosure when AI tools have been used to generate images.
2. Documentation standards: The methods, tools, and parameters used to generate images must be documented in sufficient detail to allow for reproducibility.
3. Attribution concerns: Publishers are increasingly addressing questions of attribution, copyright, and intellectual property rights related to AI-generated content.
4. Ethical considerations: Policies frequently emphasize that authors remain responsible for ensuring AI-generated content adheres to ethical standards.
5. Evolving nature: Most publishers explicitly state that their policies are subject to change as the technology and ethical understanding evolves.

Disciplinary Variations

It’s worth noting that policies often vary by discipline:

• Medical journals tend to have the strictest policies, generally limiting AI-generated content to non-critical visualizations or conceptual diagrams.
• Computer science publications often have more permissive policies, particularly for papers specifically researching AI capabilities.
• Social sciences and humanities journals are developing distinct frameworks that address the unique challenges of using AI-generated visual content in qualitative research.

Emerging Policy Consensus

As the landscape evolves, a consensus appears to be forming around these principles:

1. AI-generated images should supplement rather than replace traditional evidence
2. Complete transparency about AI use is non-negotiable
3. Authors bear ultimate responsibility for all content, regardless of how it was created
4. The technology used to create images should be documented with the same rigor as any other methodology

Acceptable vs. Unacceptable Uses of AI for Figures and Illustrations

Understanding the nuanced distinction between appropriate and inappropriate applications of AI in academic visuals is crucial for researchers navigating this rapidly evolving domain.

Acceptable Uses

Conceptual Illustrations and Diagrams

AI tools are increasingly accepted for creating conceptual illustrations that help readers understand complex ideas or processes. For example:

• Schematic representations of cellular processes
• Conceptual diagrams of theoretical frameworks
• Visual abstracts summarizing key findings
• Simplified representations of complex systems

These applications are generally accepted when:

1. They clearly serve an explanatory purpose
2. They do not claim to represent actual experimental data
3. They are properly labeled as AI-generated
4. They do not contain misleading information

Enhancing Visualization of Genuine Data

AI can be appropriately used to improve the visualization of genuine research data:

• Clarifying microscopy images through AI-assisted processing
• Creating more accessible color schemes for colorblind readers
• Generating 3D models from 2D data sets
• Improving the visual representation of statistical information

These applications are generally acceptable when:

1. The underlying data is genuine
2. The enhancement process is fully documented
3. Both original and enhanced images are available for comparison
4. The enhancement does not alter the scientific interpretation of the results

Supplementary Materials

AI-generated images often find appropriate use in supplementary materials:

• Visual abstracts for online repositories
• Alternative representations of complex data for educational purposes
• Illustrations for lay audiences or press releases
• Interactive visualizations for online versions of articles

Artistic Elements

When papers require artistic elements that do not represent scientific data, AI tools may be appropriate:

• Cover art for journals
• Decorative elements that do not convey scientific information
• Visual metaphors clearly labeled as such
• Hypothetical visualizations of concepts where no actual images exist

Unacceptable Uses

Fabrication or Simulation of Experimental Results

Using AI to create images that appear to be experimental results constitutes academic misconduct:

• Generating microscopy images when actual microscopy wasn’t performed
• Creating visualizations of experiments that were proposed but not conducted
• Producing images that appear to show statistical significance where none exists
• Generating “idealized” versions of actual experimental results

Misrepresentation of Sample Size or Variation

AI should not be used to:

• Generate multiple “examples” from a single experimental observation
• Create composite images that suggest broader sampling than occurred
• Produce “average” or “typical” specimens when limited samples were studied
• Generate visual data that masks or underrepresents natural variation

Alteration of Genuine Results

Using AI to substantively modify actual experimental images is generally unacceptable:

• Removing inconvenient artifacts or anomalies
• Enhancing patterns to appear more significant than they are
• Adding features that were not present in original data
• “Correcting” experimental results to better match hypotheses

Bypassing Ethical Approvals

AI-generated images should never be used to:

• Visualize human or animal subjects when ethical approval was not obtained
• Represent clinical outcomes that were not actually observed
• Circumvent restrictions on studies involving protected or vulnerable populations
• Create images that would otherwise require informed consent

Gray Areas Requiring Special Consideration

Several uses fall into gray areas where institutional policies and disciplinary norms may vary:

1. Predictive visualizations: AI-generated images that visualize predicted outcomes of models or simulations
2. Hypothetical mechanisms: Visualizations of theoretical mechanisms not directly observed
3. Reconstructions: Using AI to reconstruct ancient artifacts or biological structures from partial data
4. Educational simplifications: Using AI to create simplified versions of complex phenomena for teaching purposes

In these cases, authors should:

• Consult their institution’s ethics committee
• Seek guidance from journal editors prior to submission
• Provide extensive documentation of methods and limitations
• Consider including alternative visualizations using traditional methods

Alternative Approaches for Creating Compliant Visual Content

As researchers navigate the evolving landscape of AI in academic publishing, several alternative approaches can help ensure compliant, ethical, and effective visual content.

Traditional Visualization Tools with Modern Capabilities

Despite the appeal of AI-generated images, traditional visualization tools have evolved significantly and offer advantages in terms of transparency and reproducibility:

Scientific Visualization Software

• ImageJ and FIJI: Open-source platforms with extensive plugins for scientific image processing that provide full documentation of all transformations
• ParaView: Designed for data analysis and visualization with complete audit trails
• MATLAB Visualization Toolkit: Offers programmatic approaches to visualization with reproducible code
• R with ggplot2: Creates publication-quality visualizations with documented code

These tools allow for sophisticated visualizations while maintaining clear documentation trails that satisfy publisher requirements.

Vector Graphics Programs

For conceptual illustrations, vector graphics programs offer precision without the ethical complications of AI:

• Adobe Illustrator: Industry standard for creating precise scientific illustrations
• Inkscape: Free, open-source alternative that exports to journal-preferred formats
• BioRender: Specifically designed for creating scientific illustrations with pre-verified components
• Affinity Designer: Cost-effective alternative with features tailored to scientific illustration

Collaborative Approaches to Visual Content

Working with Scientific Illustrators

Professional scientific illustrators combine visual expertise with scientific understanding:

• Many institutions maintain relationships with professional illustrators familiar with publication standards
• Professional illustrators understand journal requirements and ethical considerations
• The collaborative process ensures accuracy while maintaining visual appeal
• Clear attribution and work-for-hire agreements prevent attribution issues

Interdisciplinary Collaboration

Partnering with colleagues who have visualization expertise:

• Data scientists can help develop visually compelling yet accurate data representations
• Design departments at many institutions offer collaborative opportunities
• Computer science colleagues may assist with algorithmic approaches to visualization
• Such collaborations should be acknowledged appropriately in publications

Hybrid Approaches: Supervised AI Use

When AI tools offer significant advantages, hybrid approaches can mitigate risks:

Human-in-the-Loop Methodologies

• Initial generation by AI followed by expert verification and modification
• Using AI for preliminary designs that are then finalized by human experts
• Combining AI-generated components with traditionally created elements
• Developing clear documentation protocols for these hybrid approaches

Transparent Processing Chains

• Documenting each step from raw data to final visualization
• Maintaining verifiable connections between original data and visual representations
• Providing access to intermediate steps in the visualization process
• Creating visual “audit trails” that can be included in supplementary materials

Institutional Resources and Support

Many institutions are developing resources to support compliant visual content:

Visual Ethics Committees

Some institutions have established specialized committees to:

• Review proposed visualization methodologies
• Provide guidance on appropriate use of new technologies
• Develop institutional standards for visual content
• Offer pre-submission review services

Training and Workshops

• Data visualization ethics workshops
• Training in compliant use of advanced visualization tools
• Guidance on documentation standards for visual content
• Collaborative sessions between scientists and visual design experts

Institutional Repositories and Templates

• Pre-approved templates for common visualizations
• Institutional repositories of compliant visual assets
• Standardized documentation formats for visualization methodologies
• Shared resources that have undergone ethical review

Conclusion: Navigating the Future of Academic Visuals

The landscape of AI-generated content in academic publishing continues to evolve rapidly. Researchers seeking to incorporate AI-generated visuals into their work should:

1. Stay informed: Publisher policies are frequently updated as the technology and ethical understanding evolves.
2. Prioritize transparency: When in doubt, disclose more rather than less about visualization methodologies.
3. Consider the purpose: Choose visualization methods that best serve the scientific purpose rather than simply using the newest technology.
4. Document thoroughly: Maintain comprehensive records of all processes used to create visual content.
5. Seek guidance early: Consult with editors, ethics committees, and colleagues before investing significant resources in novel visualization approaches.

As AI tools continue to advance, the academic community must collaboratively develop standards that harness their potential while preserving the integrity of scientific communication. By approaching these tools with appropriate caution and transparency, researchers can enhance the visual communication of their work while maintaining the trust that underlies scientific publishing.

The most successful researchers will be those who view AI not as a replacement for traditional visualization methods but as one tool among many—to be selected based on appropriateness for the specific scientific communication task at hand, and always used with transparency, integrity, and a commitment to scientific truth.

---

Note: This blog post reflects the current state of publisher policies as of May 2025. Given the rapidly evolving nature of both the technology and associated policies, readers are encouraged to consult the most recent guidelines from specific publishers before submission.