Skip to content
Submit incident
AII-1002-64597
ReportedReported by fewer than three independent publishers, with no official finding.Supporting sources are pending verification.

ByteDance's OmniHuman-1 generates highly realistic deepfake videos

ByteDance researchers demonstrated OmniHuman-1, an AI system that generates highly realistic deepfake videos from minimal source material. The technology raises concerns about potential misuse in political disinformation and financial fraud, reportedly contributing to billions in losses.

2 February 2025 Β· ByteDance Β· Misinformation and synthetic media
2 February 2025
AII-1002-64597
Quick facts
Category
Misinformation and synthetic media
Country
China
Date
2 February 2025
Drafted by
Raidu Incident Detection

What happened

Researchers from ByteDance demonstrated OmniHuman-1, an AI system capable of generating highly realistic deepfake videos. The system requires only a single reference image and audio to generate clips of arbitrary length, and can edit existing videos including modifying a person's limb movements.

OmniHuman-1 was trained on 19,000 hours of video content from undisclosed sources. The ByteDance team released cherry-picked samples including a fictional Taylor Swift performance, a fabricated TED Talk, and a deepfaked Einstein lecture.

Reported impact

Deepfake technology is increasingly being used in political elections and financial crimes globally. Last year, political deepfakes spread in Taiwan, Moldova, and South Africa ahead of elections, while consumers have been duped by deepfakes of celebrities promoting fraudulent investments and corporations have been swindled by deepfake impersonators.

According to Deloitte, AI-generated content contributed to more than $12 billion in fraud losses in 2023 and could reach $40 billion in the U.S. by 2027.

Index analysis

While OmniHuman-1 demonstrates technical limitations with low-quality reference images and certain poses, it represents a significant advancement in deepfake realism compared to previous techniques. The AI community tends to reverse-engineer models like these quickly, raising concerns about widespread misuse despite ByteDance not releasing the system officially.

Governance controls

Last February, hundreds in the AI community signed an open letter calling for strict deepfake regulation, though specific regulatory responses remain unclear.

Classification

Category
Misinformation and synthetic media
Industry
Media, social platforms and marketing
Country
China
Harm status
Actual harm
Intent
Unintentional
Timing
Unknown
Harm types
Economic or property, Public interest
Harmed
Business, Consumers, General public
Failure modes
Deepfake video
AI systems
OmniHuman-1

Sources and evidence

Sources not yet published

This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.

How sources are verifiedSuggest a source for this incident