Over the course of 2025, deepfakes improved dramatically. AI-generated faces, voices, and full-body performances that mimic real people advanced far beyond what even many experts expected just a few years ago. At the same time, they were increasingly used to deceive.
In many everyday settings — particularly low-resolution video calls and content circulating on social media — realism is now high enough to reliably fool nonexpert viewers. In practical terms, synthetic media have become indistinguishable from authentic recordings for ordinary users and, in some cases, even for institutions.
The surge is not limited to quality. The volume of deepfakes has grown explosively. Cybersecurity firm DeepStrike estimates an increase from roughly 500,000 online deepfakes in 2023 to about 8 million in 2025, with annual growth nearing 900%.
Several technical shifts explain this escalation. First, video realism has taken a major leap thanks to models designed to maintain temporal consistency. These systems generate coherent motion, stable identities and content that makes sense from one frame to the next. By separating information about identity from motion, the same movements can be mapped onto different people, or a single identity can perform a wide range of actions.
As a result, faces no longer suffer from the flicker, warping or distortions around the eyes and jawline that once served as reliable forensic clues.
Second, voice cloning has crossed what might be called the “indistinguishable threshold.” Just a few seconds of audio now suffice to produce a convincing clone, complete with natural intonation, emotion, pauses and even breathing sounds. This capability is already fueling large-scale fraud.
Third, consumer tools have reduced the technical barrier almost to zero. Advances such as OpenAI’s Sora 2, Google’s Veo 3 and a wave of startups mean that anyone can describe an idea, have a large language model draft a script, and generate polished audiovisual content in minutes. AI agents can automate the entire process. The ability to create coherent, storyline-driven deepfakes at scale has effectively been democratised.
This combination of volume and realism creates serious challenges for detection, especially in a media environment where attention is fragmented and content spreads faster than it can be verified. Real-world harm is already evident, from misinformation and targeted harassment to financial scams that take hold before people realise what is happening.
Looking ahead, the trajectory is clear. Deepfakes are moving toward real-time synthesis, generating live or near-live videos that closely mirror human nuance and evade existing detection systems. The frontier is shifting from static realism to temporal and behavioral coherence.
Identity modeling is converging into unified systems that capture not just how a person looks, but how they move, sound and behave over time. Entire video-call participants may soon be synthesized in real time, alongside interactive AI-driven actors and scammers deploying responsive avatars rather than fixed videos.
As the perceptual gap between synthetic and authentic media continues to narrow, defense will increasingly shift away from human judgement. Simply looking harder at pixels will no longer be enough.
Siwei Lyu is Professor of Computer Science and Engineering; Director, UB Media Forensic Lab, University at Buffalo
The Conversation