Deepfake video detection: challenges and opportunities
Open accessThe DOI matches its Crossref record; 132 of the 132 references it cites were found in a registry or an index; 43 of the 163 sources its claims cite were checked for support.
Abstract
Abstract Deepfake videos are a growing social issue. These videos are manipulated by artificial intelligence (AI) techniques (especially deep learning), an emerging societal issue. Malicious individuals misuse deepfake technologies to spread false information, such as fake images, videos, and audio. The development of convincing fake content threatens politics, security, and privacy. The majority of deepfake video detection methods are data-driven. This survey paper aims to thoroughly analyse deepfake video generation and detection. The paper’s main contribution is the classification of the many challenges encountered while detecting deepfake videos. The paper discusses data challenges such as unbalanced datasets and inadequate labelled training data. Training challenges include the need for many computational resources. It also addresses reliability challenges, including overconfidence in detection methods and emerging manipulation approaches. The research emphasises the dominance of deep learning-based methods in detecting deepfakes despite their computational efficiency and generalisation limitations. However, it also acknowledges the drawbacks of these approaches, such as their limited computing efficiency and generalisation. The research also critically evaluates deepfake datasets, emphasising the necessity for good-quality datasets to improve detection methods. The study also indicates major research gaps, guiding future deepfake detection research. This entails developing robust models for real-time detection.
Overview
CoChat Summary
## TL;DR
This is a survey of deepfake video detection that argues the field is dominated by data-driven deep learning methods but still struggles with limited labeled data, imbalance, computational cost, generalization, and adversarial robustness. The paper’s main contribution is a taxonomy of deepfake-video-detection challenges, along with datasets, methods, open issues, and future opportunities grounded in a systematic review of 132 papers.
## Key findings
- The systematic literature review extracted data from 132 papers and found a strong emphasis on deep learning methods, especially CNNs and RNNs, for deepfake video detection.
- The paper highlights several recurring problems: limited data availability, unbalanced datasets, the need for large computational resources, and the dominance of methods that are strong on accuracy but weak on real-time use and robustness.
- Reported detector results in the survey’s summary table include XceptionNet AUC-ROC 0.9965 on the DFDC challenge, EfficientNet AUC-ROC 0.9974, and some detectors with accuracy above 90%; however, the paper notes these do not capture real-world performance well.
- In the 2020 DFDC benchmark, the top-performing model achieved 82.56% accuracy on the public dataset, but only 65.18% accuracy on the black-box dataset, illustrating the gap between benchmark and harder unseen data.
## Methods
- This is a systematic literature review (SLR) focused on data-driven deepfake video detection; the paper says its main objective is to explore and analyze existing research on deepfake video detection.
- Search sources included IEEE Xplore, Google Scholar, ACM Digital Library, SpringerLink, PubMed, arXiv, CVPR, Scopus, ScienceDirect, and Web of Science; inclusion criteria restricted the review to peer-reviewed, English-language works on deepfake detection, challenges, and future research.
- The analysis synthesizes prior work into taxonomies of generation techniques, datasets, detection methods, challenges, open issues, and future opportunities rather than running a new experimental model.
## Limitations
- The paper explicitly notes that many existing detectors are evaluated mainly by accuracy, while time and cost complexity are often neglected; this is presented as a limitation of the field, not just this review.
- It states that many methods have not been evaluated on unseen deepfakes or perturbation/adversarial attacks, limiting claims of robustness and generalization.
- Evident from the text: this is a survey, so it does not provide a new unified benchmark or new experimental comparison under a common protocol; instead, it relies on the heterogeneity of the reviewed studies and datasets.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures










Verification
Claims the paper makes
34 Valid · 70 Not valid · 109 Not checked · 117 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“It has been demonstrated that GANbased synthesis approaches may create unexpectedly high-quality false videos (Karras et al. 2019).”
What the sources say
- Partly supportedcites Karras et al. 2019 sourceExperimental — accuracy not yet measured
Section 2.1 (pp. 2–3) reports that the GAN generator produces high-quality synthetic images. It does not demonstrate the creation of false videos; the video mentioned on p. 3 accompanies the image results. (LLM, conf=0.87)
states a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Diffusion models can produce more realistic images compared to GANs and VAEs (Aghasanli etal. 2023; Dhariwal and Nichol 2021).”
What the sources say
- Not checked — no full text of the sourcecites Aghasanli et al. 2023 source
- Partly supportedcites Dhariwal and Nichol 2021 sourceExperimental — accuracy not yet measured
The Conclusion states that the authors’ diffusion models achieve better sample quality than state-of-the-art GANs, supporting the GAN comparison. The Introduction discusses VAEs among likelihood-based models but does not establish a direct diffusion-versus-VAE image-quality comparison, so that part of the claim is not supported by our copy. (LLM, conf=0.85)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“It is becoming increasingly simple to use pretrained GAN to instantly replace one person's face in an image or video with the face of another person (Liu etal. 2021b).”
What the sources say
- Partly supportedcites Liu et al. 2021b sourceExperimental — accuracy not yet measured
Section 6.1 (page unavailable) describes face swapping as placing one person's face region onto another, supporting the replacement part of the claim. The Subject-agnostic section says some methods need only one image to synthesize a person's video, but it discusses reenactment rather than instant face swapping. The paper does not establish that pretrained GANs can be used to do this instantly or that doing so is becoming increasingly simple. (LLM, conf=0.88)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“The dataset contains videos with various visual artefacts and modifications, making it appropriate for testing deepfake detection algorithms (Korshunov and Marcel 2018).”
What the sources say
- Supportedcites Korshunov and Marcel 2018 source
The Deepfake Database section (pp. 2–3) describes face-swapped videos made in low- and high-quality versions using different blending techniques. The Introduction (p. 1) and Detection section (pp. 3–4) report using the dataset to evaluate deepfake detection methods, including methods based on image quality. Together, these support the claim that its visual variations make it suitable for testing detection algorithms. (LLM, conf=0.90)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Most existing deepfake detection algorithms are based on DNN because of their capability in feature extraction and selection processes (Afchar et al. 2018; Li et al. 2018; Rossler etal. 2019).”
What the sources say
- Partly supportedcites Afchar et al. 2018 sourceExperimental — accuracy not yet measured
The Proposed method (p. 3) describes this paper's deep neural network for detecting forged videos and its convolutional and pooling layers for feature extraction. It does not establish that most existing deepfake detectors use DNNs or attribute their use to feature-selection capabilities; the introduction (p. 1) says the authors knew of no other method dedicated to Deepfake detection. (LLM, conf=0.86)
- Not matchedcites Li et al. 2018
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Partly supportedcites Rossler et al. 2019 sourceExperimental — accuracy not yet measured
The Face Manipulation Methods section says more recent forensic literature concentrates on CNN-based solutions, and the Introduction credits CNNs with learning powerful image features. These passages support the prevalence of neural-network detectors and the feature-learning rationale, but they do not establish a capability in feature selection processes. (LLM, conf=0.84)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
74 more claims · 10 with their cited sources checked
Create a free account to see all 330 claims and to verify the 9 references still open.
- Probably not a substantial claim. 212
- It may state two separate claims as one. 87
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?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“these models often overfit to dataset-specific artifacts and exhibit poor generalization to unseen datasets or novel manipulation methods [13].”
From Cross-dataset video deepfake detection using Transformer and CNN architectures (2026) · cites this paper as [13]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Current survey papers [5-9, 15, 15, 16] primarily focus on reviewing detection methods based on their accuracy performance within the same dataset (within-domain evaluation), which achieves high accuracy (often exceeding 95%).”
From Deepfake detection across image, video, and audio: a comprehensive survey with empirical evaluation of generalization and robustness (2026) · cites this paper as [5]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“While current survey papers [5, 7, 9, 15] provide extensive taxonomies and theoretical discussions, they lack empirical evaluations that would enable researchers to understand the relative strengths and limitations of different approaches.”
From Deepfake detection across image, video, and audio: a comprehensive survey with empirical evaluation of generalization and robustness (2026) · cites this paper as [5]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“While surveys report impressive within-domain accuracies, they fail to examine how these same detectors perform when facing new data types, unseen manipulation techniques, or emerging generator architectures [5, 7, 9, 15].”
From Deepfake detection across image, video, and audio: a comprehensive survey with empirical evaluation of generalization and robustness (2026) · cites this paper as [5]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“While there are several surveys of passive DF detection that examine specific modalities in isolation [5-9], the field requires a holistic analysis that transcends these single approaches.”
From Deepfake detection across image, video, and audio: a comprehensive survey with empirical evaluation of generalization and robustness (2026) · cites this paper as [5]
1 more claim about this paper
Create a free account to see every claim other papers make about this one.
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 1×
“It has also been used in deepfake detection and achieved high accuracy (He et al. 2016).”
Language Models are Few-Shot Learners
Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al. (2020) · arXiv · doi:10.48550/arxiv.2005.14165 · Cited by 23,372
Generative adversarial networks
Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., & Bengio, Y. (2020) · Communications of the ACM · doi:10.1145/3422622 · Cited by 18,384
EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
Tan, M., & Le, Q. V. (2019) · arXiv · doi:10.48550/arxiv.1905.11946 · Cited by 17,072
Mastering the game of Go with deep neural networks and tree search
Silver, D., Huang, A., Maddison, C. J., Guez, A., Sifre, L., van den Driessche, G., Schrittwieser, J., Antonoglou, I., Panneershelvam, V., Lanctot, M., et al. (2016) · Nature · doi:10.1038/nature16961 · Cited by 15,612 · cited 1×
“I m p r ov e d e xp l a i n a b ilit y o f d e t ec ti on r e s u l t s (Silver et al. 2016).”
A survey on deep learning in medical image analysis
Litjens, G., Kooi, T., Bejnordi, B. E., Setio, A. A. A., Ciompi, F., Ghafoorian, M., van der Laak, J. A. W. M., van Ginneken, B., & Sánchez, C. I. (2017) · Medical Image Analysis · doi:10.1016/j.media.2017.07.005 · Cited by 15,332 · cited 1×
“However, Litjens et al. (2017) mentioned that a lack of labelled data is a common issue when applying machine learning to medical images, even when enormous volumes of unlabelled data are available (Zhang et al. 2020).”
A Style-Based Generator Architecture for Generative Adversarial Networks
Karras, T., Laine, S., & Aila, T. (2019) · 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2019.00453 · Cited by 11,714 · cited 2×
“Deepfake technology has made it possible to make these videos look real; therefore, it is necessary to assess the videos’ authenticity (Westerlund 2019; Karras et al. 2019).”
Denoising Diffusion Probabilistic Models
Ho, J., Jain, A., & Abbeel, P. (2020) · arXiv · doi:10.48550/arxiv.2006.11239 · Cited by 8,712
3D Convolutional Neural Networks for Human Action Recognition
Ji, S., Xu, W., Yang, M., & Yu, K. (2013) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2012.59 · Cited by 6,235
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2015) · arXiv · doi:10.48550/arxiv.1512.03385 · Cited by 4,196
Research works printed but not linked: 6
- Agarwal S, Farid H, Gu Y et al (2019) Protecting world leaders against deep fakes. In: CVPR workshops. pp 38–45
- Dufour N, Gully A (2019) Contributing data to deepfake detection research. Google AI Blog 1(2):3
- George AS, George AH (2023) Deepfakes: the evolution of hyper realistic media manipulation. Partn Univers Innov Res Publ 1(2):58–74
- Koopman M, Rodriguez AM, Geradts Z (2018) Detection of deepfake video manipulation. In: The 20th Irish machine vision and image processing conference (IMVIP). pp 133–136
- Le TN, Nguyen HH, Yamagishi J et al (2021) OpenForensics: large-scale challenging dataset for multi-face forgery detection and segmentation in-the-wild. In: Proceedings of the IEEE/CVF international conference on computer vision. pp 10117–10127
- Malik MH, Ghous H, Qadri S et al (2023) Frequency-based deep-fake video detection using deep learning methods. J Comput Biomed Inform 4(02):41–48
Cited by
Digital deception: generative artificial intelligence in social engineering and phishing
Schmitt, M., & Flechais, I. (2024) · Artificial Intelligence Review · doi:10.1007/s10462-024-10973-2 · Cited by 198
“Kaur et al. (2024) have published a comprehensive study on deepfake video detection, highlighting significant threats posed by false content created using generative AI.”
Enhanced deepfake detection with densenet and cross-vit
Siddiqui, F., Yang, J., Xiao, S., & Fahad, M. (2025) · Expert Systems with Applications · doi:10.1016/j.eswa.2024.126150 · Cited by 62
Babaei, R., Cheng, S. S. M., Duan, R., & Zhao, S. (2025) · Journal of Sensor and Actuator Networks · doi:10.3390/jsan14010017 · Cited by 103
“For Alzheimer’s patients, animated representations of loved ones can provide comfort and emotional connection [58].”
Deepfake video detection methods, approaches, and challenges
Alrashoud, M. (2025) · Alexandria Engineering Journal · doi:10.1016/j.aej.2025.04.007 · Cited by 49
Maheshwari, R. U., & Paulchamy, B. (2024) · Automatika · doi:10.1080/00051144.2024.2400640 · Cited by 56
Metadata
- Authors
- Kaur, A., Noori Hoshyar, A., Saikrishna, V., Firmin, S., & Xia, F.
- Affiliations
- Federation University (AU)
- RMIT University (AU)
- Corresponding author
- Kaur, Achhardeep
- Venue
- Artificial Intelligence Review
- Published
- 2024-05-29
- Volume
- 57
- Issue
- 6
- ISSN
- 0269-2821 (ISSN-L)
- 1573-7462
- OpenAlex
- W4399138486
- Cited by
- 214
- References
- 118
- Citations by year
- 2026: 93 · 2025: 109 · 2024: 12
- FWCI
- 29.81
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0 (tdm)
- creativecommons.org/licenses/by/4.0 (vor)
- Locations
- Artificial Intelligence Review (publishedVersion, cc-by)
- FedUni ResearchOnline (Federation University Australia) (submittedVersion, cc-by)
- Funders
- Federation University Australia
- 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
- deepfake video detection, deep learning, misinformation, fake content, class imbalance, labeled training data, computational requirements, deepfake datasets, real-time detection
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.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Deepfake video detection: challenges and opportunities. https://app.cochat.ai/papers/10.1007/s10462-024-10810-6
- The paper
- Kaur, A., Noori Hoshyar, A., Saikrishna, V., Firmin, S., & Xia, F. (2024). Deepfake video detection: challenges and opportunities. Artificial Intelligence Review. https://doi.org/10.1007/s10462-024-10810-6
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.