For work that will be published in or alongside a peer-reviewed paper, mostly not. Many of the leading journals do not permit AI-generated images or video in published research, and they hold human authors responsible for the accuracy of every figure.
Policies vary. Some prohibit AI-generated images outright, others allow them only with full disclosure. The safe assumption for anything destined to sit beside a paper is that AI-generated visuals are restricted unless that journal's policy says otherwise.
Last reviewed May 2026. Journal AI policies are changing quickly, so confirm the current position of your target journal before you submit.
What the leading journals say about AI-generated figures
The picture is not uniform, and the differences matter. Here is where the major publishers currently stand.
Nature
Prohibits AI-generated images. AI tools cannot be listed as authors, and any AI use must be documented in the Methods section. See the Nature Portfolio AI policy.
Springer Nature
Across its portfolio of thousands of titles, does not permit AI-generated images or video in published work. The policy names animation and 2D or 3D visual representations directly. See the Springer Nature editorial policies.
Science (AAAS)
Requires that human authors remain responsible for the accuracy and integrity of all content, including figures, and that AI use in figure creation is disclosed.
Cell Press
Requires AI-generated figures to be clearly labelled, and does not allow AI in data figures without review.
PLOS
Encourages transparency and does not impose an outright ban where AI use is properly disclosed.
The takeaway: if your visual is going into a published paper, or will be promoted alongside one, treat AI-generated imagery as off-limits until you have checked the exact policy of the journal you are submitting to.
Why are journals cautious about AI images?
The core concern is integrity, and it doesn't go away as the technology improves.
Generative AI can fabricate or distort detail in ways that look convincing but are wrong, and a figure that can't be checked against ground truth is a threat to the scientific record rather than a contribution to it. Accountability for accuracy sits with the human authors, and that responsibility can't be handed to a model. Better models don't soften this. If anything they sharpen it, because plausible-but-wrong is more dangerous than obviously wrong.
This is why the line matters. A published figure has to be verifiably correct and built by someone who can stand behind it. Atmosphere and narrative in a film carry no such burden, which is why generative tools are at home there and nowhere near a figure.
Where is AI actually appropriate in science communication?
AI is well suited to narrative, atmosphere and cinematic storytelling. It is not suited to being the source of a verifiable technical claim.
A generative tool can carry the sweep of a discovery, the scale of a system, the feeling of a piece of research. What it cannot reliably do is render the exact shape of a protein, an anatomical relationship a clinician would certify, or the precise mechanism a paper asserts, to a standard a specialist would put their name to.
A simple test sorts most cases: if a detail being wrong would damage a specialist's reputation or mislead someone about the science, it is a verifiable claim and should be built by people. If the visual carries the story without being the evidence for a fact, AI can carry it. Many of the best projects use both, with the verifiable elements hand-built and the cinematic ones generated under human direction.
What does scientific animation cost?
It depends entirely on what kind of work it is, and the range is wide.
Hand-built scientific and medical 3D animation, the accurate, certifiable kind, commonly runs into the tens of thousands of pounds for a few minutes, because it is skilled, deliberate work checked frame by frame. Fully cinematic live-action or photoreal CGI, with built environments and characters, runs into six figures for as little as two minutes of finished footage, which is why most research teams have historically had no realistic route to it at all.
Generative AI changes the second figure, not the first. It makes cinematic ambition reachable on a research budget. It does not make accurate technical animation cheaper, because that work still has to be built by hand to be trustworthy.
How to commission visuals that are safe to publish
First, what "publish" means here. We mean placing a visual in, or formally alongside, a peer-reviewed paper, the part of the record a journal controls. That's different from sharing a film publicly on your own channels, where the standard is yours to set, not the journal's.
For anything bound for the formal record, choose a studio that guarantees no AI-generated content in what it delivers, can document its process, and builds every verifiable technical element by hand. If there's no AI-generated content in the deliverable, there's nothing in it for a journal's AI policy to exclude.
What Scrolly Science makes
We do both kinds of work, and we keep them completely separate.
Accurate science visuals, built by hand
Highly technical 3D medical and molecular animation, scientific illustration, publication-ready figures, motion graphics and explainers. This work contains no AI-generated content in anything we deliver. It's built by people, frame by frame, to sit in or beside a peer-reviewed paper at the standard a specialist would certify.
Scrolly Cinematic, our AI-assisted storytelling
Cinematic films made with generative AI under full human creative direction, for the narrative, scale and atmosphere of your research. We never use Scrolly Cinematic for verifiable technical or medical detail, and never for anything destined for a published figure. Anything that has to be exact goes to the hand-built work above. The line is the whole point: AI for the story, people for the facts.