CoChat

Deepfake Attacks: Generation, Detection, Datasets, Challenges, and Research Directions

Open access

The DOI matches its Crossref record; 125 of the 126 references it cites were found in a registry or an index, and 1 are not checked yet; 3 of the 86 sources its claims cite were checked for support.

Abstract

Recent years have seen a substantial increase in interest in deepfakes, a fast-developing field at the nexus of artificial intelligence and multimedia. These artificial media creations, made possible by deep learning algorithms, allow for the manipulation and creation of digital content that is extremely realistic and challenging to identify from authentic content. Deepfakes can be used for entertainment, education, and research; however, they pose a range of significant problems across various domains, such as misinformation, political manipulation, propaganda, reputational damage, and fraud. This survey paper provides a general understanding of deepfakes and their creation; it also presents an overview of state-of-the-art detection techniques, existing datasets curated for deepfake research, as well as associated challenges and future research trends. By synthesizing existing knowledge and research, this survey aims to facilitate further advancements in deepfake detection and mitigation strategies, ultimately fostering a safer and more trustworthy digital environment.

Verified 2026-11-19 · 1 reference not checked yet

Overview

CoChat Summary

TL;DR

This survey reviews deepfake generation techniques, detection clues and deep learning architectures, datasets, and open challenges. It synthesizes state-of-the-art methods and reports benchmark performance and dataset characteristics, while noting real-time detection, generalization, interpretability, privacy, dataset quality, and adversarial perturbations as key unresolved issues.

Key findings

  • The paper identifies five primary deepfake manipulation types: face synthesis, attribute manipulation, identity swap/face swapping, face reenactment, and lip-syncing.
  • Detection approaches are organized by clues: spatial artifacts, biological/physiological signs, audio-visual inconsistencies, convolutional traces, identity information, facial emotions, temporal inconsistencies, and spatial-temporal features.
  • Reported detection performance examples include Zhao et al. with ACC = 99.8 on Celeb-DF and ACC = 92.1 on DFDC; Wang, J. et al. with ACC = 97.93 and AUC = 99.51 on FF++(HQ) and AUC = 99.80 on Celeb-DF; and Chen et al. with Celeb-DF AUC = 99.80 and FF++ AUC = 99.39.
  • Dataset examples include DFDC with over 100,000 total clips from 3426 paid actors, Celeb-DF with 590 real and 5639 fake videos, KoDF with 62,166 authentic and 175,776 fake videos, and ForgeryNet with 2.9 million images, 221,247 videos, and 15 manipulation methods.

Methods

  • Design: narrative literature survey/review; the paper states that no new data were created or analyzed.
  • Scope: covers deepfake manipulation types, generation techniques such as GANs and autoencoders, detection clues and deep learning models, existing datasets, and challenges/future research directions.
  • Analysis: synthesizes existing knowledge and tabulates recent detection models with features, techniques, datasets, and intra-dataset performance (Table 2), plus key characteristics of prominent deepfake datasets (Table 3).

Limitations

  • The paper states field-level challenges: real-time detection, generalization and robustness under compression/noise, lack of interpretability, privacy concerns, dataset quality and visual differences from online deepfakes, and adversarial perturbations.
  • Text makes evident: as a survey, it provides no new experiments or empirical validation; reported performance numbers are compiled from cited works and not independently reproduced.
  • No explicit limitations of the survey itself are stated in the paper.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/computers12100216

Cites

125 of 126 references verified0 citation errors · 0 not checkable

Statistics

Recomputing p-values needs the paper’s full text.

Figures

Figure 1 The five principal categories of deepfake manipulation.
Figure 1 The five principal categories of deepfake manipulation.CC-BY · From the PDF
Figure 2 Clues and features employed by deepfake detection models in the identification fake content. Figure 2. Clues and features employed by deepfake detection models in the identification of deepfake content.
Figure 2 Clues and features employed by deepfake detection models in the identification fake content. Figure 2. Clues and features employed by deepfake detection models in the identification of deepfake content.CC-BY · From the PDF
Figure 3 Overview of predominant deep learning architectures, networks, and frameworks employed in the development of deepfake detection models. Figure 3. Overview of predominant deep learning architectures, networks, and frameworks employed in the development of deepfake detection models.
Figure 3 Overview of predominant deep learning architectures, networks, and frameworks employed in the development of deepfake detection models. Figure 3. Overview of predominant deep learning architectures, networks, and frameworks employed in the development of deepfake detection models.CC-BY · From the PDF
Image 4
CC-BY · From the PDF
Image 5
CC-BY · From the PDF
Figure 6 Frequency of usage of different deepfake datasets in the discussed detection models within Figure 6. Frequency of usage of different deepfake datasets in the discussed detection models within this survey.
Figure 6 Frequency of usage of different deepfake datasets in the discussed detection models within Figure 6. Frequency of usage of different deepfake datasets in the discussed detection models within this survey.CC-BY · From the PDF

Verification

Claims the paper makes

3 Valid · 0 Not valid · 68 Not checked · 25 Own results and interpretations

16 more claims

Create a free account to see all 96 claims and to verify the 83 references still open.

See all claims — free account
  • Probably not a substantial claim. 66
  • It may state two separate claims as one. 29
  • The quote only partly states this claim. 1

Create a free account to see these claims.

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?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

2 of 2

Citations

Cites

Research works printed but not linked: 10

  1. Kandari, M., Tripathi, V., and Pant, B. (2023, January 15–17). A Comprehensive Review of Media Forensics and Deepfake Detection Technique. Proceedings of the 2023 10th International Conference on Computing for Sustainable Global Development (INDIACom), New Delhi, India.
  2. Prajwal, K.R., Mukhopadhyay, R., Philip, J., Jha, A., Namboodiri, V., and Jawahar, C.V. (2019). Proceedings of the 27th ACM International Conference on Multimedia, Nice, France, 15 October 2019, Association for Computing Machinery.
  3. Hu, J., Liao, X., Gao, D., Tsutsui, S., Qin, Z., and Shou, M.Z. (2023). DeepfakeMAE: Facial Part Consistency Aware Masked Autoencoder for Deepfake Video Detection. arXiv.
  4. Wang, J., Wu, Z., Ouyang, W., Han, X., Chen, J., Lim, S.-N., and Jiang, Y.-G. (2022). Proceedings of the ICMR—International Conference on Multimedia Retrieval, Newark, NJ, USA, 27–30 June 2022, Association for Computing Machinery, Inc.
  5. Cai, Z., Stefanov, K., Dhall, A., and Hayat, M. (December, January 30). Do You Really Mean That? Content Driven Audio-Visual Deepfake Dataset and Multimodal Method for Temporal Forgery Localization: Anonymous Submission Paper ID 73. Proceedings of the International Conference on Digital Image Computing: Techniques and Applications, DICTA, Sydney, Australia.
  6. Tan, M., and Le, Q.V. (2020). EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks, PMLR.
  7. Yan, Z., Sun, P., Lang, Y., Du, S., Zhang, S., and Wang, W. (2022). Landmark Enhanced Multimodal Graph Learning for Deepfake Video Detection. arXiv.
  8. Dolhansky, B., Bitton, J., Pflaum, B., Lu, J., Howes, R., Wang, M., and Ferrer, C.C. (2020). The DeepFake Detection Challenge (DFDC) Dataset. arXiv.
  9. Khalid, H., Tariq, S., and Woo, S.S. (2021). FakeAVCeleb: A Novel Audio-Video Multimodal Deepfake Dataset. arXiv.
  10. Korshunov, P., and Marcel, S. (2018). DeepFakes: A New Threat to Face Recognition? Assessment and Detection. arXiv.

Cited by

See all 117

Metadata

Authors
Naitali, A., Ridouani, M., Salahdine, F., & Kaabouch, N.
Affiliations
  • University of Hassan II Casablanca (MA)
  • University of North Carolina at Charlotte (US)
  • University of North Dakota (US)
Corresponding author
  • Naitali, Amal
  • Kaabouch, Naima
Venue
Computers
Publisher
MDPI AG
Published
2023-10-23
Volume
12
Issue
10
Pages
216
ISSN
2073-431X (ISSN-L)
OpenAlex
W4387874311
Cited by
117
References
114
Citations by year
2026: 28 · 2025: 56 · 2024: 33
FWCI
10.62
Type
journal article
Language
en
Open access
Locations
  • Computers (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
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, deepfake detection, deepfake generation, deepfake datasets, misinformation, political manipulation, propaganda, reputational damage, fraud, synthetic media

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.

A language model drew the claims from the paper on 2026-10-07, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-07 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (qwen3.8-omni-flash).

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

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.