Existing solutions are predominantly limited to single-modality analysis

The gap

Evidence profile

Sourced from the future work and stated research gap of the source papers, classified as general, spanning 3 journals.

Supporting evidence — 3 representative gaps

• Deepfake Face Detection and Recognition Systems (2026) · International Journal of Scientific Engineering and Research · doi

  1) Deepfake detection continues to develop thanks to improvements in artificial intelligence and security technologies. In particular, nowadays, researchers do not Volume 14 Issue 5, May 2026 www.ijser.in Licensed Under Creative Commons Attribution CC BY Paper ID: SE26508232113DOI: https://dx.doi.org/10.70729/SE2650823211327 of 28 International Journal of Scientific Engineering and Research (IJSER) ISSN (Online): 2347-3878 SJIF (2025): 8.036 on the generalization problem, showing that many detection models struggle when tested on unseen datasets. (ScienceDirect) [7] Bharati, N. (2025). Explainable deepfake detection framework. This research introduces explainable AI (XAI) techniques in deepfake detection, making results more interpretable for forensic and legal applications. (ScienceDirect) [8] Khan, A. A. et al. (2025). Survey on multimedia- enabled deepfake detection. The paper discusses multimodal detection systems combining video, audio, and facial cues to improve accuracy and robustness. (Springer) [9] Qureshi, S. M. et al. (2024). Survey of digital forensic methods for multimodal deepfake detection. This study explores detection across multiple media types and highlights challenges in detecting deepfake content on social media platforms. (PMC) [10] Gupta, G. (2023). Deepfake detection using multimodal and machine learning approaches. The research into datasets, benchmarks, and provides machine learning techniques used in deepfake detection systems. (MDPI) insights [11] Wang, T. (2022). Deepfake detection: A reliability- focused survey. This paper emphasizes challenges such as robustness, transferability, and interpretability in detection models. (arXiv) [12] ResearchGate Survey (2025). A survey on deepfake video detection. This review highlights the current state of deepfake detection and stresses the need for better real-world applicability and robustness. (ResearchGate) [13] Banerjee, S. (2025). Survey on deepfake detection technologies. The paper examines both traditional and modern detection methods, including biometric and AI- based techniques. (ResearchGate) [14] IJRCSEIT (2026). Deepfake detection through deep learning. This study compares different neural network architectures such as CNNs, RNNs, and transformers in terms of performance and efficiency. (IJSRCSEIT) [15] Singh, S. (2025). Integrative review of deepfake detection and multimedia forensics. The authors discuss the social and cybersecurity implications of deepfakes and suggest interdisciplinary solutions. (PMC) limit themselves to studying the number of blinks; instead, they try to develop a system capable of recognizing changes and adapting to them. it is important 2) Multimodal biometric techniques do not imply only facial features; to analyze audio characteristics in addition. Today's security technologies use not only face recognition and verification but also fingerprint, voice, and multifactor authentications. 3) Thanks to the development of real-time detection technology, it becomes possible to prevent attempts to post any deepfake videos online instantly. In addition, blockchain technology should be regarded as promising within this area. 4) Finally, the role of datasets increases since face detection models prepare fight with deepfakes. the Simultaneously, detection tools are improving, and privacy laws are introduced. for

  generalfuture work
  Keywords: detection deepfake survey techniques multimodal technologies models datasets robustness learning researchgate develop thanks security ijser
• Deepfake Media Generation and Detection in the Generative AI Era: A Survey and Outlook (2026) · ACM Computing Surveys · doi

  In this paper, we reviewed deepfake generation and detection methods, constructing a comprehen- sive taxonomy of methods across image, video, audio and multimodal domains. After discussing the methods included in our taxonomy, we turned our attention to datasets used for deepfake detection, with a particular focus on the results reported by top performing models. Moreover, we evaluated some of the best methods on our novel benchmark, BioDeepAV, aiming to assess the generalization capacity of current deepfake detectors to out-of-distribution data. The results show that the distribution gap can greatly affect state-of-the-art deepfake detectors, pinpointing the need for more robust models. 8.1 Future Perspectives As the deepfake generation technology continues to evolve, a number of complex challenges are raised, spanning from a technical nature to ethical and societal concerns. Furthermore, based on the ACM Comput. Surv., Vol. 37, No. 4, Article 111. Publication date: August 2026. Deepfake Media Generation and Detection in the Generative AI Era: A Survey and Outlook 111:25 observed gaps in deepfake literature, there are several directions which we recommend exploring in future work. We next split the discussion between deepfake generation and detection. Generation safeguarding. The widespread of publicly available generation methods, together with their increased capability, has accelerated the accessibility and realism of synthetic media, making it critical to reflect on the potential risks. First of all, researchers should develop safeguards (e.g. invisible watermarks), while the developers of open-source models should integrate such mechanisms or implement usage restrictions to mitigate misuse. In parallel with the development of generative methods, regulatory and legal frameworks need to be implemented to define and enforce boundaries of fair and safe use. Finally, but not least important, the resilience of the society against manipulated media can be boosted through public awareness campaigns and digital literacy initiatives. Cross-domain and open-set benchmarks. We consider that the most important future direction is the development of deepfake detectors that can generalize across multiple generative tools. Our new benchmark, BioDeepAV, will come in handy to test the generalization capacity of deepfake detection models in the future. Cross-domain and open-set benchmark need to be continuously developed, as generative methods get better over time. One-class learning. We foresee that the principal avenues for improving generalization in deepfake detection lie in the exploration of multimodal architectures, while shifting away from traditional supervised learning. Although, unsupervised methods have been tried in current literature, they still lag behind supervised methods. One field from which future research can draw inspiration for improvement is video anomaly detection, where models are trained exclusively on normal data on various

  generalfuture work
  Keywords: deepfake detection generation models future generative benchmark generalization detectors need media open taxonomy across video
• Explainable Multimodal AI for Deepfake Detection and Digital Content Authenticity (2026) · IJARCCE · doi

  Existing solutions are predominantly limited to single-modality analysis. Prior work has sought to close the gaps in deepfake detection, explainable AI, and multimodal misinformation analysis. The surveyed literature reveals a consistent set of limitations across the literature: single-modality coverage, lack of transparency, absence of contextual grounding, and insufficient throughput for real-time deployment.

  generalstated research gapevidence 5/5
  Keywords: existing solutions predominantly limited single-modality analysis prior work

Questions about this gap