Generating and Detecting Various Types of Fake Image and Audio Content: A Review of Modern Deep Learning Technologies and Tools

Arash Dehghani; Hossein Saberi

REVIEW 4 major objections 7 minor 30 references

Generating and Detecting Various Types of Fake Image and Audio Content: A Review of Modern Deep Learning Technologies and Tools

T0 review · 4 major / 7 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This review maps the deepfake generation-detection arms race by organizing synthetic media into four facial manipulation families plus voice conversion, each tied to a generative model family and to open-source detection tools.

desk verdict A broad but shallow deepfake survey whose tool tables are handy, but uncited accuracy claims and misattributed references make the review unreliable until fixed. read the letter →

arxiv 2501.06227 v1 pith:LO74VCUK submitted 2025-01-07 cs.CR cs.LG

Arash Dehghani , Hossein Saberi This is my paper · ORCID

classification cs.CRcs.LG

keywords DeepfakeGenerationDetectionArtificialIntelligenceDeepNeuralNetworksLearningVariationalAutoencodersGenerativeAdversarialDiffusionModels

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This review tries to give a single structured picture of the deepfake problem: how synthetic images, videos, and audio are made with modern deep learning, and how detectors try to catch them. It argues that generation has moved from traditional graphics editing to deep generative models, specifically VAEs, GANs, and diffusion models, and that this shift has lowered the skill barrier so far that nearly anyone can produce convincing fakes. The authors organize generation into four facial manipulation families, namely face swapping, reenactment, talking-face generation with lip sync, and facial attribute editing, plus voice conversion, and they organize detection into fake-face detection and broader AI-generated-content detection. If the map is right, the practical consequence is that no single detector can cover the whole threat: localized face forgeries and globally synthetic images leave different artifacts, so defenses must be matched to the generative pipeline that produced the fake.

What carries the argument

The organizing device is the four-way taxonomy of deepfake manipulation, face swapping, face reenactment, talking-face and lip-sync generation, and facial attribute editing, extended by voice conversion, with each category anchored to a generative architecture, a VAE, GAN, or diffusion model, and to a set of open-source tools. The taxonomy does the argument's work: it lets the authors match generation methods to detection methods, fake-face versus AIGC, spatial versus frequency-based, and it lets them claim that face swapping and reenactment are the highest-risk manipulations. A second load-bearing device is the GAN's adversarial training loop, the generator-versus-discriminator contest, which the paper describes as the conceptual engine common to both generation and detection.

What would settle it

A reader could settle the review's reliability by tracing every quantitative claim to its source, starting with the Section 6 assertion that CNN-LSTM hybrids exceed 95% accuracy on FaceForensics++ and Celeb-DF; if that number has no published basis, or if the barrier-to-entry claim in Section 3 is not supported by the cited source, the review's central characterization of detection performance and tool accessibility is unsupported.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is a taxonomy plus a trend claim. The current deepfake threat is defined by a handful of deep generative model families: VAEs enable face swapping by sharing an encoder between two faces, GANs drive face swapping, reenactment, and attribute editing through adversarial training, and diffusion models are taking over generation and producing outputs that differ enough from GAN-era fakes to break older detectors. The paper further claims that detection splits into fake-face detection, where artifacts are local, and AIGC detection, where artifacts are global, and that hybrid CNN-LSTM detectors can exceed 95% accuracy on benchmark datasets even while diffusion-generated content remains a moving target. The review's contribution is therefore not a new algorithm but a consolidated map of generation tools, categories, and detection strategies, presented as evidence of an arms race in which detection must keep chasing generation.

Load-bearing premise

The review's conclusions rest on the assumption that every cited source says what the text says it says and that the tool tables give a representative, consistent evaluation; in particular, Section 3 attaches the barrier-to-entry claim to the Transformer architecture paper, and Section 6 states the 95% CNN-LSTM detection accuracy without any citation.

Editorial extensions

If this is right

Detection systems must be built and evaluated separately for fake-face forgeries and for globally synthetic AIGC images, since each leaves a different type of artifact.
Because open-source tools and pre-trained models lower the skill barrier, practical defenses should expect high volumes of deepfakes produced by non-experts, not just by specialized researchers.
Face swapping and face reenactment are the two manipulation families the paper singles out as the most socially threatening, so forensic priorities should concentrate on those two.
Reported CNN-LSTM detection accuracy above 95% on standard benchmarks does not imply the same performance on diffusion-generated content, which the paper identifies as a distinct and harder case.
Voice conversion detection cannot simply borrow image-based methods, because audio deepfakes require analysis of acoustic features and speech-pattern nuances rather than visual artifacts.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Not the paper's own claim: if diffusion models continue to displace GANs as the default generator, detector accuracy trained on GAN-era benchmarks should decay measurably, which is testable by running current detectors on a diffusion-generated face dataset.
The paper's tables mix code-hosting star counts, release dates, and subjective ratings as if they were comparable performance signals; one could turn those tables into a living benchmark that re-evaluates each tool on a fixed, dated dataset.
The parallel versus non-parallel voice conversion distinction suggests a testable asymmetry: if self-supervised speech features continue to improve, non-parallel conversion quality should approach parallel quality, weakening the trade-off the paper describes.
The review's focus on facial and voice media leaves open a modular extension: synthetic body motion and full-scene video generation are likely to require detection artifacts not captured by the four-family taxonomy.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague,

This is a review, not a research paper, and its value is limited to orientation. The only original artifacts are the GitHub tool tables with star ratings. Those are handy for someone entering the area, and the chapter structure covering face swapping, reenactment, lip sync, attribute editing, voice conversion, and detection is sensible. The high-level background on neural net architectures is fine but reads like boilerplate from existing surveys.

The soft spots are real. The paper's central claim of being a comprehensive, reliable review is undercut by citation and consistency problems. Reference [10] is repeatedly cited for statements it cannot support—e.g., that deep learning lowers the barrier to entry, and that face swapping poses a significant threat. Section 6 asserts that CNN+LSTM hybrids exceed 95% accuracy on FaceForensics++/Celeb-DF with no citation; Section 5.4 likewise claims up to 95% identification failure without a source. That pattern makes the survey non-auditable. The star ratings in the tool tables have no stated methodology, so 'comprehensive' is not verifiable. The cross-reference to 'Table 2' in the voice conversion section is actually Table 3, a minor but telling slip.

These are fixable with a careful revision: add citations or remove unsupported numbers, define the tool selection and rating criteria, and correct the table references. As it stands, an informed reader can get a quick list of tools and a bird's-eye view, but should not trust the specific numbers or the attribution of claims.

I don't think this deserves a desk rejection for being a review; it deserves peer review so that a referee can chase down the citation errors and require the methodology. Without those corrections, it is not citable as a reliable survey. For a reading group, I would not bring it; there are better surveys already in its own reference list.

Referee Report

4 major / 7 minor

Summary. The paper is a survey of deepfake generation and detection, targeting audio, image, and video content. It reviews deep learning foundations (CNNs, RNNs, autoencoders, VAEs, transformers, GANs, diffusion models), categorizes deepfake types (face swapping, reenactment, lip synchronization, face attribute editing, voice conversion), and compiles tables of open-source tools for generation and detection, annotated with release dates, repositories, documentation, performance star ratings, and GitHub stars. The stated goal is to provide a comprehensive and current overview of the field and the 'arms race' between generation and detection, with emphasis on practical tools. The paper is entirely a literature and tool compilation; it does not present new experimental results or a formal evaluation methodology.

Significance. If the citation and accuracy issues were corrected, this survey would be a useful entry point for researchers and practitioners seeking an organized map of deepfake generation/detection tools and methods. Its strengths are the broad taxonomic coverage, the inclusion of very recent (2023-2024) open-source tools with repository links, and the parallel treatment of generation and detection. However, the value of a review of this kind rests entirely on the traceability of its factual claims and the reproducibility of its tool assessments. The current manuscript contains uncited quantitative claims, misattributed references, and an undocumented star-rating scheme, so the central claim of being a comprehensive and reliable review is not yet substantiated. The paper makes no novel technical contribution, but a well-executed survey would still be a legitimate contribution to this rapidly moving area.

major comments (4)

[Section 6 and Section 5.4] Section 6 (first paragraph) states without any citation that 'hybrid approaches combining CNNs with LSTMs can achieve detection accuracy exceeding 95% on benchmark datasets like FaceForensics++ and Celeb-DF.' Section 5.4 similarly states that manipulations such as digital reshaping or beautification 'can, in some cases, lead to identification failures of up to 95%.' These are specific quantitative claims with no supporting reference, model specification, or evaluation protocol. A survey must make every attributed or numerical claim checkable; these two claims are load-bearing for the paper's assertions about detection effectiveness and biometric vulnerability, and they should either be removed or supported by a cited study that describes the exact model, dataset, and protocol.
[Section 3, Section 5.1, and Section 5.2] Several references do not support the claims to which they are attached. Reference [10] (Vaswani et al., 'Attention is all you need') is cited in Section 3 for the claim that deep learning techniques 'significantly lower the barrier to entry for creating convincing deepfakes' and in Section 5.1 for the claim that face swapping and face reenactment 'present a significant threat to society'; the Transformer paper contains neither claim. Section 5.2 states that 'Reference [9] is a significant and commonly used resource for facial expression transfer,' but reference [9] is a survey paper on deepfake generation and detection, not a facial-expression-transfer dataset or tool. These misattributions directly undermine the verifiability of the survey's factual content and need to be corrected with appropriate sources.
[Section 5.5 and Section 6] The table cross-references are inconsistent and will mislead readers. Section 5.5 says 'Table 2 presents a list of all the relevant collected tools' when discussing voice conversion, but the voice-conversion tools actually appear in Table 3. The same section refers to 'the model presented by tool number one in Table 1' in the voice-conversion context, but Table 1 lists image/video tools. Section 6 states 'Table 3 lists the key tools employed in this research' for deepfake detection, but the detection tools are in Table 4. All table references need to be rechecked and corrected so that each described tool category points to the correct table.
[Tables 1-4] The 'Performance' column in Tables 1-4 uses a star rating (zero to five stars), but the manuscript never describes the evaluation methodology behind these ratings. Section 5.5 asserts that 'the ordering in Table 1 reflects the results of our evaluations,' yet no criteria, dataset, or procedure is given. Without a stated methodology, the star ratings are not reproducible and cannot be considered part of a comprehensive, objective tool comparison. The authors should either provide a clear evaluation protocol (including how tools were selected, what dimensions were scored, and who performed the scoring) or relabel the column as a subjective/community-based indicator with an explicit caveat.

minor comments (7)

[Author affiliation] The email address 'hsaberi @ihu.ac.ir' contains an erroneous space before '@'; it should read 'hsaberi@ihu.ac.ir'.
[Abstract] There is a typo in the abstract: 'potential threats p osed' should be 'potential threats posed'.
[Figure 2] The text references 'Figure 2 provides a structured overview of deepfake types' in Section 5, but no actual figure is included in the manuscript; a placeholder or the figure itself must be supplied.
[Figure 1 caption] The caption reads 'Demonstration The structure and training methods of GANs'; this should be grammatically corrected, for example to 'Demonstration of the structure and training methods of GANs'.
[Section 1 and Section 2] The paper uses 'Chapter three' and 'Chapter four' in Section 1, but the manuscript is organized into sections; the terminology should be consistent (e.g., 'Section 3' and 'Section 4').
[Section 5.5] The paragraph discussing the k-nearest-neighbor voice conversion method is attached to a table row ('tool number one in Table 1'), but the correct table is Table 3; in addition, the relationship between this detailed method description and the table format is not explained for readers who are not familiar with the cited tools.
[Abstract and Keywords] The heading 'Key Words' should be 'Keywords' to match standard journal style; the keyword list could also include 'audio deepfake' and 'voice conversion' for better indexing.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the review compiles external sources, and its citation-accuracy issues are verifiability defects, not circularity.

full rationale

This paper is a narrative review with no mathematical derivation, fitted parameters, or novel predictive claims. Its content is a compilation of external references and tool tables, so there is no derivation chain whose outputs could reduce to its inputs. The concern raised by the reader—uncited performance numbers (e.g., the >95% CNN-LSTM accuracy claim in Section 6) and a misattributed reference (Vaswani et al. cited for lowering the barrier to entry)—is a defect in evidence quality and verifiability, not circularity. No step in the paper defines a concept in terms of the thing it is supposed to explain, fits a parameter and then calls the result a prediction, or imports a conclusion from the authors' own prior work. There are no self-citations at all. Therefore, applying the hard rule that circularity must be exhibited by a specific reduction, no circular step is present. The appropriate score is 0.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

No new entities are introduced; the paper is a review. The main assumptions are that the cited literature is used accurately and that the tool tables reflect a consistent evaluation, but both assumptions are strained by the errors noted above.

assumptions (2)

domain assumption The cited references accurately support the statements attributed to them
The review's claims rely on secondary citations; several are misattributed, e.g., ref [10] (Vaswani et al.) is cited for deepfake tool accessibility.
ad hoc to paper The tool tables and star ratings reflect a consistent, reproducible evaluation methodology
No methodology for assigning performance ratings is described.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Generating and Detecting Various Types of Fake Image and Audio Content: A Review of Modern Deep Learning Technologies and Tools." pith.science (2026). https://pith.science/paper/LO74VCUK

@misc{pith2026250106227,
  author       = {Pith},
  title        = {Pith review of: Generating and Detecting Various Types of Fake Image and Audio Content: A Review of Modern Deep Learning Technologies and Tools},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LO74VCUK}},
  note         = {Machine review of arXiv:2501.06227}
}

read the original abstract

This paper reviews the state-of-the-art in deepfake generation and detection, focusing on modern deep learning technologies and tools based on the latest scientific advancements. The rise of deepfakes, leveraging techniques like Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), Diffusion models and other generative models, presents significant threats to privacy, security, and democracy. This fake media can deceive individuals, discredit real people and organizations, facilitate blackmail, and even threaten the integrity of legal, political, and social systems. Therefore, finding appropriate solutions to counter the potential threats posed by this technology is essential. We explore various deepfake methods, including face swapping, voice conversion, reenactment and lip synchronization, highlighting their applications in both benign and malicious contexts. The review critically examines the ongoing "arms race" between deepfake generation and detection, analyzing the challenges in identifying manipulated contents. By examining current methods and highlighting future research directions, this paper contributes to a crucial understanding of this rapidly evolving field and the urgent need for robust detection strategies to counter the misuse of this powerful technology. While focusing primarily on audio, image, and video domains, this study allows the reader to easily grasp the latest advancements in deepfake generation and detection.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 22 canonical work pages

[10]

Attention is all you need,

A. Vaswani, "Attention is all you need," Advances in Neural Information Processing Systems, 2017

2017
[9]

Deepfake generation and detection: Case study and challenges,

Y. Patel et al., "Deepfake generation and detection: Case study and challenges," IEEE Access, 2023

work page 2023
[1]

Comparison of deepfake detection techniques through deep learning,

M. Taeb and H. Chi, "Comparison of deepfake detection techniques through deep learning," Journal of Cybersecurity and Privacy, vol. 2, no. 1, pp. 89-106, 2022

work page 2022
[2]

Deep learning for deepfakes creation and detection: A survey,

T. T. Nguyen et al., "Deep learning for deepfakes creation and detection: A survey," Computer Vision and Image Understanding, vol. 223, p. 103525, 2022

work page 2022
[3]

Deepfake attacks: Generation, detection, datasets, challenges, and research directions,

A. Naitali, M. Ridouani, F. Salahdine, and N. Kaabouch, "Deepfake attacks: Generation, detection, datasets, challenges, and research directions," Computers, vol. 12, no. 10, p. 216, 2023

work page 2023
[4]

DeepFaceLab: Integrated, flexible and extensible face-swapping framework,

I. Perov et al., "DeepFaceLab: Integrated, flexible and extensible face-swapping framework," arXiv preprint arXiv:2005.05535, 2020

arXiv 2005
[5]

Deepfake generation and detection: A benchmark and survey,

G. Pei et al. , "Deepfake generation and detection: A benchmark and survey," arXiv preprint arXiv:2403.17881, 2024

arXiv 2024
[6]

Understanding the Impact of AI -Generated Deepfakes on Public Opinion, Political Discourse, and Personal Security in Social Media,

P. L. Kharvi, "Understanding the Impact of AI -Generated Deepfakes on Public Opinion, Political Discourse, and Personal Security in Social Media," IEEE Security & Privacy, 2024

work page 2024

Show all 30 references

[7]

Deepfake video detection: challenges and opportunities,

A. Kaur, A. Noori Hoshyar, V. Saikrishna, S. Firmin, and F. Xia, "Deepfake video detection: challenges and opportunities," Artificial Intelligence Review, vol. 57, no. 6, pp. 1-47, 2024

2024
[8]

Simswap: An efficient framework for high fidelity face swapping,

R. Chen, X. Chen, B. Ni, and Y. Ge, "Simswap: An efficient framework for high fidelity face swapping," in Proceedings of the 28th ACM international conference on multimedia , 2020, pp. 2003-2011

2020
[11]

Brahme, Comprehensive biomedical physics

A. Brahme, Comprehensive biomedical physics. Newnes, 2014

2014
[12]

Pattern recognition and machine learning,

C. M. Bishop, "Pattern recognition and machine learning," Springer google schola, vol. 2, pp. 1122- 1128, 2006

2006
[13]

Deep learning,

Y. LeCun, Y. Bengio, and G. Hinton, "Deep learning," nature, vol. 521, no. 7553, pp. 436-444, 2015

2015
[14]

Imagenet classification with deep convolutional neural networks,

A. Krizhevsky, I. Sutskever, and G. E. Hinton, "Imagenet classification with deep convolutional neural networks," Advances in neural information processing systems, vol. 25, 2012

2012
[15]

Backpropagation applied to handwritten zip code recognition,

Y. LeCun et al. , "Backpropagation applied to handwritten zip code recognition," Neural computation, vol. 1, no. 4, pp. 541-551, 1989

1989
[16]

Neural network model for a mechanism of pattern recognition unaffected by shift in position-neocognitron,

K. Fukushima, "Neural network model for a mechanism of pattern recognition unaffected by shift in position-neocognitron," IEICE Technical Report, A, vol. 62, no. 10, pp. 658-665, 1979

1979
[17]

Phoneme Recognition Using Time -Delay Neural Network,

W. Alexander, "Phoneme Recognition Using Time -Delay Neural Network," IEEE transactions on acoustics, speech, and signal processing, 1989

1989
[18]

Neural networks and physical systems with emergent collective computational abilities,

J. J. Hopfield, "Neural networks and physical systems with emergent collective computational abilities," Proceedings of the national academy of sciences, vol. 79, no. 8, pp. 2554-2558, 1982

1982
[19]

On the properties of neural machine translation: Encoder -decoder approaches,

K. Cho, "On the properties of neural machine translation: Encoder -decoder approaches," arXiv preprint arXiv:1409.1259, 2014

2014 arXiv
[20]

A comprehensive overview of Deepfake: Generation, detection, datasets, and opportunities,

J. W. Seow, M. K. Lim, R. C. Phan, and J. K. Liu, "A comprehensive overview of Deepfake: Generation, detection, datasets, and opportunities," Neurocomputing, vol. 513, pp. 351 -371, 2022

2022
[21]

Fine -grained face swapping via regional gan inversion,

Z. Liu et al. , "Fine -grained face swapping via regional gan inversion," in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 8578-8587

2023
[22]

ShapeEditor: A stylegan encoder for stable and high fidelity face swapping,

S. Yang et al. , "ShapeEditor: A stylegan encoder for stable and high fidelity face swapping," Frontiers in Neurorobotics, vol. 15, p. 785808, 2022

2022
[23]

The creation and detection of deepfakes: A survey,

Y. Mirsky and W. Lee, "The creation and detection of deepfakes: A survey," ACM computing surveys (CSUR), vol. 54, no. 1, pp. 1-41, 2021

2021
[24]

Denoising diffusion probabilistic models,

J. Ho, A. Jain, and P. Abbeel, "Denoising diffusion probabilistic models," Advances in neural information processing systems, vol. 33, pp. 6840-6851, 2020

2020
[25]

Exposing Lip-syncing Deepfakes from Mouth Inconsistencies,

S. K. Datta, S. Jia, and S. Lyu, "Exposing Lip-syncing Deepfakes from Mouth Inconsistencies," arXiv preprint arXiv:2401.10113, 2024

2024 arXiv
[26]

Designing a New Non -parallel Training Method to Voice Conversion with Better Performance than Parallel Training,

M. Ghorbandoost and V. Saba, "Designing a New Non -parallel Training Method to Voice Conversion with Better Performance than Parallel Training," Paramedical Sciences and Military Health, vol. 10, no. 2, pp. 6-16, 2015

2015
[27]

Parallel vs. Non -Parallel Voice Conversion for Esophageal Speech,

L. Serrano, S. Raman, D. Tavarez, E. Navas, and I. Hernaez, "Parallel vs. Non -Parallel Voice Conversion for Esophageal Speech," in INTERSPEECH, 2019, pp. 4549-4553

2019
[28]

Voice conversion with just nearest neighbors,

M. Baas, B. van Niekerk, and H. Kamper, "Voice conversion with just nearest neighbors," arXiv preprint arXiv:2305.18975, 2023

2023 arXiv
[29]

Wavlm: Large-scale self-supervised pre-training for full stack speech processing,

S. Chen et al., "Wavlm: Large-scale self-supervised pre-training for full stack speech processing," IEEE Journal of Selected Topics in Signal Processing, vol. 16, no. 6, pp. 1505-1518, 2022

2022
[30]

Hifi-gan: Generative adversarial networks for efficient and high fidelity speech synthesis,

J. Kong, J. Kim, and J. Bae, "Hifi-gan: Generative adversarial networks for efficient and high fidelity speech synthesis," Advances in neural information processing systems, vol. 33, pp. 17022-17033, 2020

2020

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [10]

Attention is all you need,

A. Vaswani, "Attention is all you need," Advances in Neural Information Processing Systems, 2017

2017

[2] [9]

Deepfake generation and detection: Case study and challenges,

Y. Patel et al., "Deepfake generation and detection: Case study and challenges," IEEE Access, 2023

work page 2023

[3] [1]

Comparison of deepfake detection techniques through deep learning,

M. Taeb and H. Chi, "Comparison of deepfake detection techniques through deep learning," Journal of Cybersecurity and Privacy, vol. 2, no. 1, pp. 89-106, 2022

work page 2022

[4] [2]

Deep learning for deepfakes creation and detection: A survey,

T. T. Nguyen et al., "Deep learning for deepfakes creation and detection: A survey," Computer Vision and Image Understanding, vol. 223, p. 103525, 2022

work page 2022

[5] [3]

Deepfake attacks: Generation, detection, datasets, challenges, and research directions,

A. Naitali, M. Ridouani, F. Salahdine, and N. Kaabouch, "Deepfake attacks: Generation, detection, datasets, challenges, and research directions," Computers, vol. 12, no. 10, p. 216, 2023

work page 2023

[6] [4]

DeepFaceLab: Integrated, flexible and extensible face-swapping framework,

I. Perov et al., "DeepFaceLab: Integrated, flexible and extensible face-swapping framework," arXiv preprint arXiv:2005.05535, 2020

arXiv 2005

[7] [5]

Deepfake generation and detection: A benchmark and survey,

G. Pei et al. , "Deepfake generation and detection: A benchmark and survey," arXiv preprint arXiv:2403.17881, 2024

arXiv 2024

[8] [6]

Understanding the Impact of AI -Generated Deepfakes on Public Opinion, Political Discourse, and Personal Security in Social Media,

P. L. Kharvi, "Understanding the Impact of AI -Generated Deepfakes on Public Opinion, Political Discourse, and Personal Security in Social Media," IEEE Security & Privacy, 2024

work page 2024