Compare Nature’s AI risk framework, Science’s permission requirement, and eLife’s disclosure rules and limits before preparing a journal figure.
SciFig Team
SciFig editorial team
You have spent three weeks on a set of experiments. The data is solid. The story is clear. Now you need figures — and you have heard that AI tools can produce publication-quality scientific illustrations in minutes. There is just one problem: you do not know whether your target journal will accept them, flag them, or penalize you for using them at all.
For submission, match each figure’s type and actual AI-assisted steps to the target journal’s current instructions. This guide helps you identify the relevant image requirements, prepare a description of your workflow, and place the disclosure where the journal specifies.
This SciFig guide covers the current stances of Nature Publishing Group, Science (AAAS), and eLife, notes where Cell Press could not be verified, identifies what is universally off-limits, and gives you practical templates for citing AI tools correctly in your manuscripts.
Note: Journal policies evolve. You can use each journal's author information pages as the reference for the latest wording; the Nature Portfolio, Science and eLife positions below are quoted from the publishers' own policy pages; the Cell Press section is flagged where it could not be verified.
Describe AI Use by the Steps Used to Create the Figure
For submission, describe the actual steps used for each figure and where AI took part. Conceptual illustrations, plots from original data, and workflow diagrams use processes suited to their content.
Comparison of AI-generated artwork and AI-assisted image processing (conceptual figure generated with SciFig)
SciFig AI Illustration (AI scientific illustration) creates mechanism and concept images from descriptions, sketches, or reference material, and you can convert any result to editable SVG to modify every element. SciFig FlowChart (flowcharts and diagrams) outputs native editable SVG workflows, and SciFig DataChart (data-driven charts) generates measured results precisely from your original table.
SciFig AI Illustration uses AI image models to generate artwork. SciFig DataChart uses AI to plan the chart and produce plotting code, then runs Python against your original table to calculate and draw the result. SciFig FlowChart uses AI to generate and revise SVG diagrams. Manual SVG editing and export use the existing figure. Record the steps actually used in the location specified by the journal.
Do not use generative AI to alter microscopy, gels, or other primary research images. Keep the source and the processing record, and follow the target journal’s image-integrity and disclosure requirements.
Nature Publishing Group Policies
Nature Publishing Group (NPG), which publishes Nature, Nature Medicine, Nature Methods, Nature Communications, and dozens of other journals, established AI content policies in early 2023 and replaced them wholesale during 2026. If you are working from guidance written before that — including earlier versions of this page — the sentence you remember, that Springer Nature journals are "unable to permit" AI-generated images, is no longer on the policy page, and neither are the three exceptions that used to accompany it. What replaced it is not a relaxation so much as a different axis: use is now judged by risk level rather than by tool type, and the obligations attach to accountability and disclosure.
Nature policy guide
The operative sentence in Springer Nature's editorial policy now reads:
"Use of AI is governed according to risk level, not by tool type. The same framework applies across all roles: authors, reviewers and editors."
The core requirements across NPG titles are:
1. There is no blanket ban on AI-generated figures — and no blanket permission. Use is sorted into three tiers: Green (assistive, "without influencing scientific, scholarly or evaluative judgement"), Amber (may influence "interpretation, framing, emphasis, or evaluative judgement, but remains under human control"), and Red (not permitted).
Nature Portfolio assesses AI use by its contribution and risk. Fabricated results, undisclosed substitution for scholarly judgment, and prohibited photorealistic imagery remain unacceptable. For an explanatory pathway diagram, record its purpose and the AI generation and editing steps used, and prepare the figure according to the journal’s current requirements.
Retain source materials and record the steps used to make the figure. Editable SVG lets you revise labels, shapes, and connections.
4. AI cannot be listed as an author. Consistent across all publishers, and now explicitly in the Red tier ("Assigning authorship or accountability to AI systems or tools"). AI tools do not meet authorship criteria — they cannot take responsibility for the work, respond to correspondence, or provide consent.
5. The text side got stricter, not looser. The old policy said in as many words that "AI assisted copy editing" did not need to be declared. That sentence is gone. Language polishing now sits in Green, where "Disclosure enhances trust and transparency," and "extensive copy editing or writing support" has moved up to Amber, which is "permitted with human oversight, verification, and transparency through disclosure." If you learned the permissive version, you learned a rule that no longer exists.
Two practical notes. The Portfolio-level policy states principles; the journal-level author guidance is where the mechanics live, and it is more specific — LLM use "should be properly documented in the Methods section (and if a Methods section is not available, in a suitable alternative part) of the manuscript." And the policy page carries no version number and no effective date, while stating that these policies "will be reviewed periodically." The journal's current page is the most up-to-date reference, and you can read it alongside this summary.
Science (AAAS) Policies
Science magazine and its family of journals (Science Advances, Science Translational Medicine, Science Immunology) require explicit editorial permission before any AI-generated image can appear — a default of "no" that can be lifted case by case, rather than a disclosure-based default of "yes."
"AI-generated images and other multimedia are not permitted in the Science journals without explicit permission from the editors."
What this means in practice:
Permission comes first, disclosure second. You cannot simply generate an illustrative figure, disclose it, and submit. Exceptions "may be granted in certain situations — e.g., for images and/or videos in manuscripts specifically about AI and/or machine learning," are "evaluated on a case-by-case basis," and "should be disclosed at the time of submission."
Where AI use is disclosed is unusually specific. Authors using AI-assisted technologies "as components of their research study or as aids in the writing or presentation of the manuscript should note this in the cover letter and in the methods section or acknowledgments section." The cover letter requirement is easy to miss and is not how most journals handle it.
AI cannot be an author or a cited source. AI-assisted technologies "do not meet the Science journals' criteria for authorship," and sources cited in Science content may not be authored or co-authored by AI tools.
There is a stated consequence. "Editors may decline to move forward with manuscripts if AI is used inappropriately."
Reviewers are restricted too. Reviewers may not enter any part of a manuscript into an LLM, as that would breach confidentiality, though they may use AI to revise their own writing if the system does not train on inputs and the use is declared.
Science also notes that its "position on AI-generated multimedia may change with the evolution of copyright law and industry standards" — again signalling that copyright, not accuracy, is the governing concern.
Cell Press Policies
Cell Press, the Elsevier-owned publisher of Cell, Molecular Cell, Cell Reports, and related journals, operates under both Elsevier's broader AI policy framework and Cell Press-specific editorial guidelines.
Elsevier distinguishes explanatory diagrams from data visualizations. Its generative AI policy permits explanatory flowcharts, decision trees, timelines, and experimental workflows when authors verify them and disclose AI use in the figure caption and the manuscript’s AI declaration. Data plots must come directly from original data through reproducible computational or statistical methods; report the methods and the tool, model, version, and developer in Methods. Generative alteration of primary research images remains prohibited. Check the individual Cell Press journal’s instructions as well.
Cell Press editorial workflow: manuscript with AI figures → AI-disclosure review checkpoint with 5 criteria → accepted (green) or revision needed (orange) (Figure generated with SciFig)
Elsevier: Transparency, author responsibility, and appropriate use:
Transparency. Authors must declare AI tool use in a dedicated section of the manuscript, typically following the acknowledgments. The declaration should specify the tool, the version or access date, and the purpose of use.
Responsibility. Authors remain fully accountable for the content of AI-generated or AI-assisted figures. Cell Press journals expect authors to be able to explain and defend any figure regardless of how it was produced.
No AI authorship. Consistent with industry norms, Cell Press journals do not permit AI tools to be listed as authors or co-authors.
Cell Press journals apply particular scrutiny to image integrity. The journals use automated image analysis tools to screen submissions for inappropriate manipulation. While these tools were developed with traditional image fraud in mind, they can also flag inconsistencies introduced by generative AI. Because editable figures let you change any element, you can use the integrity screening as a reference and adjust anything that might read as an unintended artifact.
Apply the publisher’s conditions to the actual figure. Describe its purpose and the AI generation and editing steps used in the location specified by the journal, and check the target Cell Press journal’s additional instructions.
See AI Scientific Figure Generation in Action
Watch how researchers create publication-ready scientific figures from text descriptions.
Across Nature Publishing Group, Science, and eLife, the policies diverge sharply on images while converging on text. Assuming a policy transfers between publishers is the most expensive mistake available in this area.
Journal policy comparison
Policy Element
Nature Publishing Group
Science (AAAS)
eLife
AI-generated illustrative figures
No blanket ban — judged by risk tier, disclosure expected
Not permitted without explicit editor permission
General article-content AI clause requires Methods disclosure; image-type rules also apply
Copyright + industry standards; position may change
Author accountability, with separate image-integrity and striking-image rules
Disclosure location
Methods section, per journal-level guidance
Cover letter and methods or acknowledgments
Materials and Methods
AI authorship
Not permitted
Not permitted
Not permitted
Stated consequence
—
Editors may decline to proceed
— (general integrity policies apply)
Generative AI for manuscript text
Permitted with disclosure and human accountability
Permitted with disclosure
Permitted; describe in Materials and Methods
Hard ban on one specific image
—
—
Striking image (CC-BY licensing)
Assess the same pathway schematic under each destination’s rules. At eLife, apply the general AI disclosure clause together with image-type requirements; a Nature-family journal applies its risk framework and journal-specific instructions, while Science requires prior editorial permission for AI-generated images. If you may transfer a manuscript, check every journal in the target set — see eLife AI policy for the distinction between stated rules and interpretation.
Permission, disclosure, and source verification are separate requirements. Science requires editorial permission for generated images. eLife requires AI disclosure, preserves experimental-image integrity rules, and prohibits AI-generated striking images. Springer Nature applies a risk framework with human accountability and transparency. Keep records of the actual workflow, then apply the target journal’s conditions. For tool workflows, see the 2026 illustration tools comparison.
Smaller society journals and specialty publications are less standardized. Some have explicit AI policies, many do not. When a policy is absent, the default expectation at most journals is that images represent what they purport to represent — which means using AI to create content presented as experimental evidence remains problematic regardless of whether a policy explicitly addresses it.
The Red Line — What Is Still Banned
Publisher positions on illustrative figures vary widely, as the table above shows. But several categories of AI use remain universally prohibited, at permissive and restrictive journals alike.
Fabrication of experimental evidence. Using a generative AI tool to create an image — a gel band, a microscopy field, a histological section — that is then presented as an actual experimental result is scientific fraud. It does not matter that the image was AI-generated rather than hand-drawn; the deception is identical. This is not a gray area. The retraction and potential misconduct consequences are severe.
Undisclosed AI use. Using an AI tool to generate figures and not disclosing it is a policy violation at every major publisher that has issued AI guidelines. Even if the scientific figures are scientifically accurate, failing to disclose violates the transparency requirements that underpin research integrity.
Misleading data visualization. Using AI (or any tool) to create visual representations that misrepresent statistical results, exaggerate effect sizes, or present data in ways that lead to incorrect conclusions is prohibited. This applies whether the distortion is intentional or inadvertent — researchers are responsible for ensuring their visualizations accurately represent their data.
Experimental-image processing that changes the evidence. Retain original acquisition files and follow the target journal’s image-integrity rules for any permitted adjustment. Uniform application and disclosure do not by themselves authorize AI enhancement or upscaling. Do not introduce, remove, or invent features, or make ambiguous results appear more conclusive.
Generation of figures for fraudulent purposes. Using AI to generate figures for fabricated studies, duplicate publications, or papers where the described experiments were not conducted is covered by existing misconduct frameworks and is treated identically to traditional fabrication.
The shared concern in this list is deception and misrepresentation. Permission for an accurate, disclosed explanatory figure is a separate question. Apply eLife’s general AI clause together with its image-type rules, check the risk framework and journal instructions at Springer Nature, and obtain explicit editorial permission where Science requires it.
How to Cite AI Tools in Your Paper
Getting the disclosure right is as important as using the tools honestly. Here are practical templates for the two most common locations: the Methods section and the Acknowledgments section.
Manuscript guide showing 3 highlighted sections for AI tool citation: Methods (full citation), Acknowledgments (secondary mention), Figure legend (disclosure) (Figure generated with SciFig)
Methods Section — General Template
AI-generated figures in this manuscript were produced using [Tool Name, version/access date]. Figures [X, Y] are AI-generated conceptual illustrations. All AI-generated figures were reviewed by the authors for scientific accuracy and are intended to illustrate [mechanism/process/experimental design], not to represent experimental data.
Methods Section — Specific Example
Figure 3 was prepared using SciFig (scifig.ai), with [workspace/model or version and its contribution]. The figure shows [a conceptual illustration, a workflow diagram, or results plotted from the original table]. The authors made [actual edits] using [source material].
Acknowledgments Section — Short Form
AI tools: [Tool Name] was used for figure generation. [Tool Name] was not involved in study design, data collection, analysis, or interpretation, and does not qualify for authorship.
Acknowledgments Section — Elsevier/Cell Press Format
Declaration of generative AI and AI-assisted technologies in the writing process: During the preparation of this work, the authors used [Tool Name] for figure generation. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.
A few practical notes on citations:
Never list the AI tool as an author. Use it in the methods or acknowledgments as a tool, not a contributor.
Record the tool and the model or software version when available, plus the access date where relevant. For SciFig, you can cite the platform (SciFig 3.0, scifig.ai) and the workspace used — SciFig AI Illustration, SciFig DataChart or SciFig FlowChart — following the journal's disclosure template.
Be specific about which figures were AI-generated. "Figures 1A and 3C were AI-generated" is better than a vague general statement.
State what the scientific figures represent — clarifying that they are conceptual illustrations rather than experimental evidence removes ambiguity.
Check journal-specific format requirements. Some journals have standardized the AI disclosure statement format; use their template if one is provided.
Create Scientific Figures Now
Describe your scientific figure in natural language — get publication-ready illustrations in minutes.
Journal AI policies are actively evolving. The positions described here reflect publicly available guidelines as of early 2026; the specific journal's author guidelines page is the reference for the current wording, and the editorial office can answer case-specific questions.
Frequently Asked Questions
Choose the preparation method separately for each figure, with the target journal's rules as the reference. For conceptual and illustrative figures, eLife’s general AI clause requires Materials and Methods disclosure and its image rules still apply; Science requires explicit editorial permission for AI-generated images; Springer Nature applies a risk framework and journal-specific requirements. For numerical results, SciFig DataChart (data-driven charts) generates the chart precisely from your original table: AI plans the chart and plotting code, then Python calculates and renders it from the supplied data, and you keep the data, settings, and code. For experimental images, keep the acquired images and permitted processing records. Never substitute generated microscopy, gels, histology, or invented measurements for real evidence. For conceptual drafting, see the illustration workflow guide.
Failure to disclose AI use is a policy violation at journals that require disclosure. Consequences range from requiring a correction (if the scientific figures are otherwise legitimate) to retraction if the non-disclosure is considered part of a broader pattern of misrepresentation. In cases of honest omission where the scientific figures are accurate and properly produced, journals typically allow authors to add a disclosure statement via correction. A short disclosure statement keeps your submission straightforward, and the templates above make it quick to add.
Check the journal’s rules for the specific operation. AI upscaling or cleanup of conceptual artwork may require disclosure even when it changes no scientific claim. For experimental images, retain the original acquisition files and follow the permitted image-processing methods; disclosure alone does not authorize a prohibited enhancement. Record the tool and operation and use the required disclosure location.
Reviewers check whether a figure accurately represents the biology, chemistry, or mechanism described and whether its preparation follows the journal’s rules. With editable figures you can adjust any label or relationship directly, keep your sources, and add the disclosure or editorial permission the journal asks for.
Fewer than there were. Science journals still do not permit AI-generated images "without explicit permission from the editors" — a default of no, liftable case by case. Springer Nature, which used to state it was "unable to permit" them, replaced that prohibition during 2026 with a risk-tier framework: the only image-specific ban left is on "Creating photorealistic images (deepfakes)," while everything else turns on disclosure and on accountability remaining human. Two things follow. A blanket ban is no longer the industry direction, and a policy summary older than a few months may be describing a policy that no longer exists, including at journals not covered here.
Sources
Nature Portfolio AI editorial policy — Springer Nature's risk-assessment framework, its three tiers, and the core expectations that apply to authors
Science journals editorial policies — the "Artificial intelligence (AI)" section, covering AI-generated images, disclosure locations, and reviewer restrictions
Boundaries. The Nature Portfolio and Science statements quoted above come from the publishers' own policy pages. The Cell Press section could not be verified because cell.com blocks automated access, and is flagged as unverified in place. Publisher AI policies are changing quickly; the journal's own author guidelines are the reference for the current text. See our editorial standards for how these guides are researched and corrected.
Keep going
Recommended resources
Templates, tools, and further reading for your research
