Venus-DefakerOne2 reports new scores across image, video and spoof forensics
Original title: Venus-DefakerOne2: A Vision-Forensics Foundation Model with Broader Coverage
Authors report 96.7% on DeepFake faces and 96.0% on AI-generated videos after training on tens of millions of samples.
Venus-DefakerOne2 is a vision-forensics model meant to cover digital manipulation, physical spoofing, images and video in one binary-decision system. Tide-Style-Tuning alternates LoRA and full-parameter fine-tuning, and Entropy-based Test Adaptation picks the prediction modality from prediction entropy. The authors say it beats specialized expert models and its predecessor, DefakerOne, on the listed benchmarks.
Unconfirmed: Benchmark scores and the state-of-the-art claim are the authors' reported results, not an independent evaluation.
- Reported scores include 93.3% on AI-generated images and 87.6% on tampered documents.
- Manipulated natural images score 89.6% and physical spoofs 93.1% in the paper.
- Training uses tens of millions of open-source and in-house samples.
- The model is positioned for binary forensic decisions, not generation.
- AI models
- Venus-DefakerOne2DefakerOne
From the original article
1 min read at huggingface.co
Real-world forgery is inherently heterogeneous: it manifests not only as digital manipulation but also as physical spoofing, and is carried not only by images but also by videos. Despite substantial progress, a unified anti-forgery foundation model covering such a broad threat spectrum remains absent.
By Chenfan Qu, Yachao Liang, Mingqi Fang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.