Can Deepfakes be created on a whim?
Open accessAbstract
Recent advancements in machine learning and computer vision have led to the proliferation of Deepfakes. As technology democratizes over time, there is an increasing fear that novice users can create Deepfakes, to discredit others and undermine public discourse. In this paper, we conduct user studies to understand whether participants with advanced computer skills and varying level of computer science expertise can create Deepfakes of a person saying a target statement using limited media files. We conduct two studies; in the first study (n = 39) participants try creating a target Deepfake in a constrained time frame using any tool they desire. In the second study (n = 29) participants use pre-specified deep learning based tools to create the same Deepfake. We find that for the first study, of the participants successfully created complete Deepfakes with audio and video, whereas for the second user study, of the participants were successful in stitching target speech to the target video. We further use Deepfake detection software tools as well as human examiner-based analysis, to classify the successfully generated Deepfake outputs as fake, suspicious, or real. The software detector classified of the Deepfakes as fake, whereas the human examiners classified of the videos as fake. We conclude that creating Deepfakes is a simple enough task for a novice user given adequate tools and time; however, the resulting Deepfakes are not sufficiently real-looking and are unable to completely fool detection software as well as human examiners.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
One study asked participants to generate their own deepfakes.
“One study asked participants to generate their own deepfakes [22].”
From Human Performance in Deepfake Detection: A Systematic Review (2025) · cites this paper as [22]
Video imperfections in authentic videos can impair human judgment, leading detectors to miscategorize authentic videos as fake.
“However, some studies also showed that video imperfections (rather than computer-generated artifacts) can impair human judgment, leading detectors to miscategorize authentic videos as fake [2, 20, 22, 27].”
From Human Performance in Deepfake Detection: A Systematic Review (2025) · cites this paper as [22]
Higher-quality deepfakes are associated with fewer visual artifacts, making them more difficult to detect.
“Higher-quality deepfakes are associated with fewer visual artifacts, making them more di ffi cult to detect [20, 35, 36, 42].”
From Human Performance in Deepfake Detection: A Systematic Review (2025) · cites this paper as [22]
Mehta et al. found that one-third (34.76%) of audio-only deepfakes were perceived as somewhat convincing or vaguely like impersonation, but none of the combined audiovisual deepfakes were considered realistic enough to be believable.
“Mehta et al. [22] found that one-third of audio-only deepfakes were perceived as somewhat convincing but none of the combined audiovisual deepfakes were considered realistic enough to be believable.”
From Human Performance in Deepfake Detection: A Systematic Review (2025) · cites this paper as [22]
Imperfections in authentic videos (e.g., lighting inconsistencies or unusual framing) can lead viewers to miscategorize genuine content as manipulated.
“Second, while lower-quality deepfakes are generally easier to detect [20, 35, 36, 42], imperfections in authentic videos (e.g., lighting inconsistencies or unusual framing) can lead viewers to miscategorize genuine content as manipulated [2, 22, 27, 35, 37].”
From Human Performance in Deepfake Detection: A Systematic Review (2025) · cites this paper as [22]
Type of stimuli (video versus audio) demonstrated mixed results.
“Type of stimuli (video versus audio) and the technique used for deepfake generation (e.g., face swapping, lip-syncing, and level of morphing) demonstrated mixed results [7, 20, 22, 28, 30, 42].”
From Human Performance in Deepfake Detection: A Systematic Review (2025) · cites this paper as [22]
The technique used for deepfake generation demonstrated mixed results.
“Type of stimuli (video versus audio) and the technique used for deepfake generation (e.g., face swapping, lip-syncing, and level of morphing) demonstrated mixed results [7, 20, 22, 28, 30, 42].”
From Human Performance in Deepfake Detection: A Systematic Review (2025) · cites this paper as [22]
Video features such as visual artifacts can impact deepfake detection rates.
“Thus, video features such as these can impact detection rates [2, 20, 22, 27, 35].”
From Human Performance in Deepfake Detection: A Systematic Review (2025) · cites this paper as [22]
Cites
Deepfakes and beyond: A Survey of face manipulation and fake detection
Tolosana, R., Vera-Rodriguez, R., Fierrez, J., Morales, A., & Ortega-Garcia, J. (2020) · Information Fusion · doi:10.1016/j.inffus.2020.06.014 · Cited by 1,237 · see citing works
The Creation and Detection of Deepfakes: A Survey
Mirsky, Y., & Lee, W. (2020) · arXiv · doi:10.48550/arxiv.2004.11138 · Cited by 863
The Creation and Detection of Deepfakes: A Survey
Mirsky, Y., & Lee, W. (2022) · ACM Computing Surveys · doi:10.1145/3425780 · Cited by 863 · see citing works
Deepfake detection by human crowds, machines, and machine-informed crowds
Groh, M., Epstein, Z., Firestone, C., & Picard, R. (2022) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2110013119 · Cited by 248 · see citing works
Fooled twice – people cannot detect deepfakes but think they can
Köbis, N., Doležalová, B., & Soraperra, I. (2021) · doi:10.2139/ssrn.3832978 · Cited by 194 · see citing works
Fooled twice: People cannot detect deepfakes but think they can
Köbis, N. C., Doležalová, B., & Soraperra, I. (2021) · iScience · doi:10.1016/j.isci.2021.103364 · Cited by 193 · see citing works
Flagged first, then by citations
Cited by
Sorted by citations
Human Performance in Deepfake Detection: A Systematic Review
Somoray, K., Miller, D. J., & Holmes, M. (2025) · Human Behavior and Emerging Technologies · doi:10.1155/hbe2/1833228 · Cited by 19 · see citing works
- Authors
- Mehta, P., Jagatap, G., Gallagher, K., Timmerman, B., Deb, P., Garg, S., Greenstadt, R., & Dolan-Gavitt, B.
- Affiliations
- New York University (US)
- Laboratório para a Ciência da Computação e Informática (PT)
- ORCID
- Venue
- Companion Proceedings of the ACM Web Conference 2023
- Publisher
- ACM
- Published
- 2023-04-30
- Pages
- 1324-1334
- OpenAlex
- W4367318734
- Cited by
- 5
- References
- 6
- Citations by year
- 2025: 3 · 2024: 1 · 2023: 1
- FWCI
- 0.54
- Type
- conference paper
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- International World Wide Web Conference (WWW) (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- National Science Foundation (2016061)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Digital Media Forensic Detection
- Social Sciences › Sociology and Political Science › Misinformation and Its Impacts
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Generative Adversarial Networks and Image Synthesis
- Keywords
- deepfakes, novice users, audio-visual alignment, deepfake detection, disinformation
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
