Integrity Guide

SciViD

Visual integrity and AI use

SciViD video papers combine recorded data, model output and explanatory visuals. Viewers must always be able to tell which is which. This guide sets out how visuals must be labeled, how AI tools may be used, and what constitutes misconduct. It applies to all visuals, whether produced with AI tools, animation software or by hand.

1. Label every segment by what it shows

Every segment of a video paper must carry an on-screen label stating its status. The video description and transcript may add detail but do not replace the on-screen label.

Recorded Data: footage, micrographs, sensor data or graphs recorded during the research.
Simulation Output: results generated by a numerical or analytical model.
Model-Predicted Output: predictions generated by a trained machine-learning model.
Conceptual Animation: an explanatory animation that does not represent data.
Illustrative Metaphor: a visual analogy used to explain a concept.
Hypothetical Scenario: an imagined setup, process or outcome.

2. Requirements by paper category

Experimental video papers present original experiments performed by the authors.

Results must be shown as Recorded Data. Conceptual animations may supplement the data and must be labeled.
Adjustments of brightness, contrast or color must be applied uniformly to the whole image or sequence and must not remove or add features.
Scale bars, timestamps and playback speed (for time-lapse or slowed footage) must be shown.
AI denoising, upscaling, frame interpolation, inpainting and generative enhancement must not be applied to Recorded Data. Transitions must not blend frames of data.
AI may be used for titles, layout, narration, captions, translation and cleaning up narration audio.

Computational papers present original numerical, analytical or machine-learning research.

Model results must be labeled Simulation Output or Model-Predicted Output and must not be altered by AI upscaling, frame interpolation or generative enhancement.
Detailed requirements are set out in Publishing Computational Models and Simulations.

Review and conceptual papers synthesize existing literature.

Authors may use diagrams, animations and visual metaphors, including AI-generated ones. All must be labeled and must not be presented as evidence.

3. Core principles

No fabrication. Regardless of the tool used, it is misconduct to create visuals that suggest data or trends not present in the data, to generate missing experimental steps or results, to present stock or third-party footage as original data, to alter published figures from other authors to support a different conclusion, or to show an imagined setup or result without the “Hypothetical Scenario” label.
Consistency with sources and data. Illustrative visuals must be consistent with the data and cited sources. A process described as slow and gradual must not be animated as instantaneous.
Accuracy over spectacle. Visuals must represent scale, timescale and structure faithfully. False color must be identified as such, and objects must not be given misleading appearance or behavior.
Rights and consent. Authors must hold the rights to publish all content under SciViD’s license, including AI-generated material subject to the tool’s terms of use. Synthetic replicas of a real person’s face or voice require that person’s documented consent.
Full disclosure. Every submission must include an AI Use Statement listing:
- Tools: each AI tool used, with its version where available;
- Scope: which segments and tasks involved AI, including data analysis, literature review, script drafting, narration, translation and visuals;
- Intent: whether AI was used to present authentic content, to create illustrative material, or to generate a model’s output.

4. Verification and consequences

Editors may request raw footage, project files, simulation data or code at any stage. Violations may lead to rejection, correction or withdrawal, as set out in SciViD’s editorial policies.

5. Errors, corrections and versions

A preprint is posted so that it can be discussed and corrected. An honest error that the authors correct openly is part of normal scientific practice and is handled by a new version. Fabrication, falsification or plagiarism is misconduct and is handled under section 4.

New versions. Authors correct an error by submitting a new version with a revision note of up to 100 words stating what changed and why. New versions undergo editorial screening.
Earlier versions remain available. Each earlier version keeps its manuscript, video abstract and, where applicable, interactive lab, preserved as submitted. It is marked “Superseded by version n” and links to the current version.
Error notices. If an earlier version contains a substantive error in its results, model, video or lab, that version carries a notice stating the error in one sentence and the version that corrects it.
Withdrawal. An item may be withdrawn for misconduct, for a serious error the authors cannot correct, or for a legal or consent issue. A notice page remains, giving the title, authors, posting date and reason for withdrawal.
External records. A DOI record in an external repository is permanent and is governed by that repository’s rules; SciViD cannot remove it. Authors are advised to deposit only after editorial screening, once the version is stable.

Submission checklist

Does every segment carry an on-screen label stating its status?
Is all Recorded Data free of AI denoising, upscaling, interpolation and generative enhancement?
Are scale bars, timestamps and playback speeds shown where relevant?
Are all illustrative visuals consistent with the data and cited sources?
Is the AI Use Statement included in the submission metadata and the video credits?
Do I hold the rights and any consents needed to publish all content under SciViD’s license?
If the item includes an interactive lab, is it labeled as a Quantitative or an Illustrative model, as set out in Publishing Computational Models and Simulations?

SciViD’s rule: every visual must show what it claims to show and state what it is.

Questions about visual integrity: editor@scivid.net

Ethical standards: responsible use of generative AI

SciViD posts video preprints from researchers and, through Emerging Scholars, from students; neither pathway is peer reviewed. Generative AI tools are widely used to produce images, animation, narration and text. These standards set out how authors, reviewers and editors may use them. Labeling of visuals and the AI Use Statement are specified in SciViD’s Integrity Guide.

1. Disclosure

All AI use must be disclosed in the AI Use Statement specified in the Integrity Guide and included in the submission metadata and the video credits.

Example: “Conceptual animations in segments 3 and 5 were generated with an image-generation model and are labeled Conceptual Animation. Narration was synthesized from a script written by the authors. A large language model was used to edit the script for clarity; all scientific content was written and checked by the authors.”

2. Authorship and accountability

AI tools cannot be listed as authors.
The research question, methods, data analysis and conclusions must be the work of the named authors, who are responsible for the accuracy of all content, including content produced with AI.
Student authors must carry out and understand the work they present. Work substantially performed or written by a mentor may not be presented as the student’s own.
Contributions from mentors, teachers and other collaborators, including those provided through paid programs, must be described in a contribution statement.
AI must not be used to paraphrase the work of others without citation or to evade similarity checks.

3. Privacy and consent

People who can be identified in footage or audio must have given their consent; for people under 18, the consent of a parent or guardian is required.
Identifiable footage, voice recordings or other personal data of other people, particularly minors, must not be uploaded to third-party AI tools without the consent of the people concerned.
AI-generated content must not misrepresent or defame individuals or institutions.

4. Content and safety

SciViD does not publish content that promotes violence, hatred or illegal activity, or that depicts people or groups in a stereotyped or demeaning way.
Videos of hazardous procedures must show appropriate safety measures. SciViD may decline content whose demonstration of methods poses a significant risk of misuse.
Editors will screen submissions against these standards.

5. Use of AI by reviewers and editors

Reviewers and editors must not upload submitted manuscripts, videos, data or review reports to generative AI tools, as this breaches the confidentiality of review.
SciViD will state in its editorial policies any AI tools it uses to screen submissions.

6. Appropriate uses

AI tools can help authors to:

illustrate processes that cannot be filmed, such as molecular interactions or field distributions, using labeled conceptual animations;
produce captions, subtitles and translations;
improve the language of scripts written by the authors.

AI must not determine research results, analyses or conclusions.

7. School-aged authors

For school-aged authors, these standards draw on the Australian Framework for Generative Artificial Intelligence in Schools, published by Australian Education Ministers in 2023 and reviewed annually, in particular its emphasis on fairness, privacy and the critical use of AI tools.

8. Consequences

Violations may lead to rejection, correction or retraction. In serious cases, SciViD may inform the author’s institution or, for school students, their school. Cases will be handled in line with established publication-ethics guidance, such as that of the Committee on Publication Ethics (COPE).

Label	Meaning	Minimum evidence
Validated	The method reproduces an independent reference, and the result is converged	Benchmark comparison, convergence or repeatability test, documented source data
Simplified estimate	An order-of-magnitude or reduced-model result with stated assumptions	Assumptions listed, validity range stated, no claim beyond that range
Conceptual	An illustration of an idea, mechanism or scenario	Clearly marked as illustrative; not used as evidence

Mistake	How to detect it	How to handle it
Model used outside its regime (e.g., continuum flow where the gas is rarefied, inertia included at very low Reynolds number, Brownian motion neglected for sub-micrometer particles)	Compute the relevant dimensionless numbers (Reynolds, Knudsen, Péclet, Biot) for your parameters	State the regime explicitly; choose a model valid in it; cite the source that defines the validity range
Unit and dimension errors (mixing µm and m, °C and K, per-minute and per-second rates)	Check that every equation is dimensionally consistent; run one case by hand	Work in SI units internally; convert only for display; label every axis and value with units
Wrong or unstated boundary and initial conditions	Ask: what happens at each wall, inlet, outlet and at t = 0?	Write them out in the methods; justify each against the physical system
No comparison with a known solution	Ask: has this code ever reproduced a result I did not produce myself?	Reproduce an analytical limit or a published benchmark first; show the comparison
No convergence test	Results change when the mesh, time step or sample size is refined	Refine at least twice; report the change in the key quantity; use the converged setting
AI-generated code accepted without checking	The author cannot explain each line or each parameter	Read and comment the code; test it on a case with a known answer; record what the AI tool produced in the AI-use log
Parameter values without a source	A value appears in a table or narration with no reference	Give a reference, a measurement or an explicit assumption for every parameter
Too few samples for stochastic results	Brownian or Monte Carlo results vary noticeably between runs	Run multiple independent realizations; report the mean and spread

Mistake	Example	Correct handling
Overclaiming	“This device could revolutionize drug delivery” from a single simulation	State what was shown, under which assumptions, and what remains untested
Correlation presented as mechanism	Two quantities change together, so one is said to cause the other	Say “is associated with”; propose the mechanism as a hypothesis and state how to test it
Estimate presented as validated	An order-of-magnitude force estimate narrated as a measured result	Use the correct status label in the text, on screen and in the narration
Missing uncertainty	A single number with no error bar or range	Report the spread across runs, parameter sensitivity, or the range of validity
Selected parameters	Only the parameter set that “works” is shown	Show the parameter sweep, including cases where the effect disappears
Novelty not checked	“For the first time” without a literature search	Search the literature; cite prior work; describe precisely what is new
Extrapolation beyond the model	A result at the micrometer scale is narrated as applying to the human body	Restrict claims to the scale and conditions modeled

Mistake	Why it matters	Correct handling
Conceptual animation mistaken for evidence	A 3D render or generative-video shot looks like a result	Mark such shots on screen as “Conceptual illustration”; keep data-derived visuals visually distinct
Generative video showing physically impossible behavior	AI video tools do not obey physics; motion, scale and timing may be wrong	Use generative footage only for non-evidentiary context; never to show a result
Plots without units, scale bars or error bars	The viewer cannot judge magnitude or reliability	Label axes with quantity and SI unit; add scale bars to images; show uncertainty
Misleading color maps	Rainbow-type maps create false boundaries and are hard to read for colorblind viewers	Use perceptually uniform maps (e.g., viridis, cividis); show the color bar with units
Truncated or inconsistent axes	Small differences look large; panels cannot be compared	Start axes at zero where magnitudes are compared; use the same scale across panels
Narration diverging from the data	The voiceover states a value or trend the figure does not show	Script the narration from the results file; check every number spoken against the source
Time and length scales not shown	Slowed or accelerated motion, or magnified objects, mislead viewers	Display time stamps, playback speed and scale bars
Too much on screen at once	Viewers cannot follow equations, plots and narration simultaneously	One idea per scene; give equations time to be read

Topic	Resource	What it gives you
Reproducible computation	Sandve et al. (2013), Ten Simple Rules for Reproducible Computational Research, PLOS Comput. Biol. 9, e1003285	Practical rules: track how each result was produced, record versions and random seeds, store the raw data behind plots
Figures	Rougier, Droettboom and Bourne (2014), Ten Simple Rules for Better Figures, PLOS Comput. Biol. 10, e1003833	Designing figures for a given audience, message and medium
Color maps	Crameri, Shephard and Heron (2020), The misuse of colour in science communication, Nat. Commun. 11, 5444	Why rainbow and red–green maps distort data; how to choose accessible, perceptually uniform maps
Units	BIPM, The International System of Units (SI Brochure, 9th edition)	Authoritative definitions and rules for writing SI units and quantities
AI and authorship	COPE position statement: Authorship and AI tools	Why AI tools cannot be authors and what authors must disclose
Coding skills	Software Carpentry lessons	Free introductory lessons on the Unix shell, Git, and Python or R
Archiving data and code	Zenodo	A free general-purpose repository operated by CERN that assigns a DOI to each deposit

SciViD

Integrity Guide

Common Mistakes in Video Preprints

1. Purpose and scope

Result status labels

2. Common mistakes in modeling and computation

3. Common mistakes in interpretation and claims

4. Common mistakes in visualization and video

5. Less obvious mistakes

6. Integrity, authorship and AI disclosure

7. Pre-submission checklist

Model and computation

Results and claims

Video

Integrity and reproducibility

8. Where to find help