Yes — for some figure types, at some publishers, and almost never without disclosure. In 2026 major publishers split images into illustrative or conceptual figures (schematics, flow charts, process diagrams) and primary research images (micrographs, gels, blots, radiology, clinical photos). Generative AI that invents or alters primary experimental data is widely forbidden. Disclosed AI help for explanatory diagrams is allowed at Elsevier, Taylor & Francis and Wiley; Science journals still need explicit editor permission for AI-generated multimedia; Springer Nature’s risk framework blocks opaque generative visuals without verifiable inputs. Elsevier also bars general-purpose generative AI tools for graphical abstracts. This guide compares the publisher pages we opened on 6 November 2026. For copyright, DPI and formats, see Can AI-Generated Figures Be Used in Academic Journals?.
Why journals split “illustrative AI” from “data AI”
Publishers are not arguing about whether software can draw a box and an arrow. They are arguing about whether a reader can trust that a figure still represents the experiment.
Illustrative / conceptual figures explain a design, a pathway idea, a workflow, or a teaching concept. They are not presented as raw observations. Elsevier calls these “explanatory images”; Taylor & Francis groups them under conceptual illustrations, process flow diagrams and teaching illustrations; Wiley talks about illustrations and data visualisations with disclosure.
Primary research images are treated as evidence. Microscopy, histology, Western blots, gels, radiology scans and patient images sit in this bucket. Across the policies we checked, generative AI must not create or alter those images so that they look like experimental results.
A third, narrower category is data visualisation (plots, charts, heatmaps). Several publishers allow AI help only when the picture is faithfully derived from your own data by reproducible methods — not when a model invents a curve.
FigPad’s positioning matches the illustrative bucket only: scientific schematics and graphical abstracts you can still edit, not AI-rewritten primary data. If a reviewer needs a gel panel changed, that change has to come from the original acquisition pipeline, not from an image model.
2026 publisher policy matrix (checked 2026-11-06)
Every row below is taken from a primary publisher page opened on 2026-11-06. Journal-level instructions can be stricter or lag behind the publisher page — always re-check your target journal’s guide for authors before you submit.
| Publisher | Illustrative / conceptual figures | Primary research images (gels, micrographs, clinical) | Graphical abstracts / cover art notes | Where to disclose | Primary source (checked 2026-11-06) |
|---|---|---|---|---|---|
| Elsevier | Permitted for explanatory images (flow charts, schematics, workflows, etc.) with human oversight | Not permitted to create or alter primary research images | General-purpose generative AI image tools must not be used for graphical abstracts; use dedicated scientific illustration tools (Elsevier names BioRender and Mind the Graph as examples) or other professional illustration tools. Cover art only with prior editor/publisher permission | Figure caption (tool, version, how used) and the manuscript AI declaration | Generative AI policies for journals (policy updated June 2026) |
| Taylor & Francis | Permitted for conceptual / process / teaching illustrations; keep prompts and versions; share on request | Generative AI creating or changing images of scientific/clinical research outputs is strictly prohibited | Graphical abstracts are listed among permitted conceptual/illustration uses on the AI manuscript page — still disclose in the legend | Figure legend (tool name + version) and Declaration of AI use | Images and figures; Using AI in research and manuscript preparations |
| Science / AAAS | AI-generated images and other multimedia are not permitted without explicit editor permission | Same restriction; image-integrity rules still apply | Exceptions evaluated case by case (e.g. manuscripts specifically about AI/ML); disclose at submission | Cover letter plus Methods or Acknowledgments, per use case; see AAAS AI usage guidelines | Science Journals: Editorial Policies |
| Springer Nature / Nature Portfolio | Risk framework: AI visual content only where inputs are transparent, lawful, and verifiable; opaque generative visuals without verifiable source material are not permitted. Nature Portfolio’s AI page lists creating photorealistic images (deepfakes) as not permitted | Fabricating or misrepresenting research images remains out of bounds under integrity rules | Treat individual journal artwork guides as potentially stricter; re-check the journal | AI declaration in the manuscript; caption/legend for AI-assisted visuals | Nature Portfolio AI editorial policies; Springer Nature: AI use in manuscript preparation |
| Wiley | Permitted for illustrations / conceptual diagrams with disclosure and author verification | AI-edited photographs not permitted; “factual and evidential images” must not be generated, modified or enhanced with AI | Prefer specialised scientific illustration tools for complex realistic systems; verify licensing | Methods and/or figure captions (tool, version, date/year, AI role, author role) | Wiley AI guidelines for researchers |
| ICMJE (cross-cutting) | Does not invent a per-image whitelist; requires disclosure of AI-assisted technologies (including image creators) at submission | Authors remain responsible for accuracy, integrity and originality of AI-assisted text and images; AI cannot be an author | Follow the destination journal’s image rules | Cover letter and the appropriate manuscript section | ICMJE: Use of AI by Authors |
How to read the Springer Nature / Nature row. On 2026-11-06 the Nature Portfolio AI page presented Springer Nature’s green / amber / red risk framework rather than the older one-line ban. The companion Springer Nature “AI use in manuscript preparation” page is more explicit about figures: AI may create or enhance visual content when the underlying data or source material can be inspected and attributed; AI-generated visuals without verifiable inputs are treated as opaque and not permitted. That is a narrower door than Elsevier’s explanatory-image allowance. If you are submitting to a Nature Portfolio journal, do not assume that a Midjourney-style schematic is fine because another publisher allows conceptual AI art — check both the portfolio AI page and the journal’s figure guide (Nature figure guidelines summarise the artwork side).
Elsevier graphical abstracts — verified. The content-plan claim that Elsevier bars general-purpose AI tools for graphical abstracts is accurate on the live policy page. Elsevier’s FAQ states that authors should use dedicated scientific illustration tools (examples given: BioRender, Mind the Graph) or other professional illustration tools, and that general-purpose generative AI tools must not be used to create graphical abstracts. Licensing on free tiers of illustration tools can still block journal use — Elsevier warns authors to check publication rights.
Cell Press journals sit inside Elsevier’s publishing family. We could not load Cell Press’s own figure-guidelines page today (HTTP 403 from our fetch environment), so we do not invent a Cell-only AI table row. Default to Elsevier’s generative AI policies plus the specific Cell journal’s guide for authors. For Cell artwork rejection patterns unrelated to AI, see Cell Press figure guidelines 2026.
What “disclosure” usually means in practice
Policies agree on the substance even when the form fields differ:
- Name the tool and version (and often the developer).
- Say what it did — drafted a schematic, suggested layout, produced a first pass you then edited.
- Say what you did — verified labels against the Methods, redrew inaccurate icons, confirmed nothing looks like primary data.
- Put it where the reader looks — figure caption or legend for that panel, plus a manuscript-level AI declaration (Elsevier’s recommended heading is “Declaration of generative AI and AI-assisted technologies in the manuscript preparation process”).
- Keep records — Taylor & Francis explicitly asks authors to retain image versions and prompts for possible editorial request.
A caption pattern that matches Elsevier’s own examples looks like this:
Figure 1. Schematic overview of study design. Initial draft generated using [TOOL / VERSION] and subsequently edited by the authors. All elements were reviewed for accuracy and completeness.
Do not list the AI tool as an author. ICMJE, Elsevier, Science and Springer Nature all treat authorship as a human responsibility.
Decision checklist before you generate a figure
Use this as a pre-flight, not as a substitute for the journal page.
- Is this figure evidence or explanation? If it is a gel, micrograph, scan or clinical photo, do not run it through a generative image model.
- Who is the publisher? If Science/AAAS: obtain editor permission before relying on AI-generated multimedia. If Elsevier: explanatory figures OK with disclosure; graphical abstracts need a dedicated scientific illustration workflow, not a general image model. If Springer Nature / Nature Portfolio: only proceed when the visual can be tied to verifiable author-controlled inputs — opaque prompt-to-pretty-picture outputs are out.
- Will a reviewer ask you to change one label? Prefer an editable scientific illustration workflow (layered SVG / editable objects) over a flat PNG you cannot revise without regenerating. That is an operational issue as much as a policy one — see journal figure DPI requirements and Elsevier artwork requirements.
- Can you publish the file under the tool’s terms? Check commercial / journal publication rights before you submit.
- Can you disclose without embarrassment? If the honest caption would surprise a co-author, fix the workflow first.
Where FigPad fits (and where it does not)
FigPad is built for illustrative scientific figures: pathways, workflows, multi-panel schematics and graphical abstracts you can still edit after generation. Layered SVG, editable PPTX and PNG exports are available, and a publication certificate is included on every plan including free — details on pricing. That combination matters for the disclosure-friendly publishers above: you can name the tool, keep an editable source, and revise a label without inventing a new “primary” image.
FigPad is the wrong tool when the panel is primary experimental data. Do not ask any generative model — including ours — to invent, clean, or “improve” a blot, gel or micrograph so that it looks more publishable. Image integrity policies will treat that as misconduct regardless of which brand of model you used.
If you are producing an Elsevier graphical abstract, treat FigPad as a dedicated scientific illustration workflow rather than a general-purpose consumer image generator: keep human review, check licensing, disclose the tool in the caption, and confirm the target journal has not added a stricter local rule.
To start from a description rather than a blank canvas, use the AI scientific figure generator.
FAQ
Does “AI allowed for explanatory figures” mean Midjourney is fine for my Nature paper?
Not automatically. Elsevier’s explanatory-image allowance does not travel to every publisher. Science journals require editor permission for AI-generated multimedia. Springer Nature’s manuscript-preparation policy rejects opaque generative visuals that lack verifiable source material. Read the Nature Portfolio / Springer Nature pages and your journal’s artwork guide before you assume a general image model is acceptable.
Can I use AI to adjust brightness on a Western blot?
No under the policies we checked for primary research images. Elsevier states AI must not create or alter primary research images, and notes that brightness, contrast or colour-balance adjustments for those images belong in established image-processing software under normal integrity rules — not generative AI. Taylor & Francis prohibits generative AI for creating or changing images of research or clinical outputs. Wiley forbids AI-edited photographs and AI generation/modification of factual evidential images.
If Elsevier forbids general-purpose AI for graphical abstracts, is a scientific illustration product allowed?
Elsevier’s live FAQ says authors should use dedicated scientific illustration tools (it gives BioRender and Mind the Graph as examples) or other professional illustration tools, and must not use general-purpose generative AI image tools for graphical abstracts. Dedicated tools still need publication-capable licensing and accurate content. Disclose the tool in the caption.
Do I need to disclose AI if I only used it to brainstorm a figure I later redrew by hand?
Wiley explicitly treats exploration that never appears in the manuscript differently from visuals that ship in the paper. If the published figure contains no AI-generated pixels or layout, many disclosure rules for images may not apply — but if AI drafted manuscript text around the figure, the text rules still might. When unsure, disclose; ICMJE frames nondisclosure as a potential integrity issue.
Will meeting DPI and format rules make an AI figure acceptable?
No. DPI, column width and file format are necessary but separate from AI policy. A perfect 300 dpi TIFF can still be rejected for undisclosed AI use or for being primary-data fabrication. Handle policy and artwork specs as two checklists: this article for policy; our DPI, Nature and Elsevier artwork guides for the file.
Key takeaways
- In 2026 the decisive split is illustrative figures versus primary research images, not “AI vs no AI.”
- Elsevier allows disclosed AI for explanatory figures; bans AI on primary research images; and bars general-purpose generative AI for graphical abstracts in favour of dedicated scientific illustration tools (checked 2026-11-06).
- Taylor & Francis and Wiley allow disclosed conceptual illustrations / visualisations and forbid generative AI on research or evidential images.
- Science/AAAS requires explicit editor permission before AI-generated multimedia.
- Springer Nature / Nature Portfolio currently use a risk framework that rejects opaque generative visuals without verifiable inputs — do not import Elsevier’s explanatory-image rule by analogy.
- Disclose tool, version and human verification in the caption and the manuscript AI statement; never list AI as an author.
- Keep an editable source file so policy-compliant figures remain revisable after peer review.
Policies change. Every claim in the matrix above is tied to a URL we opened on 2026-11-06. Re-check your target journal the week you submit.
