Pith. sign in

REVIEW 11 cited by

Towards Universal Fake Image Detectors that Generalize Across Generative Models

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2302.10174 v2 pith:ZUCMAEB2 submitted 2023-02-20 cs.CV cs.LG

Towards Universal Fake Image Detectors that Generalize Across Generative Models

Utkarsh Ojha , Yuheng Li , Yong Jae Lee This is my paper
classification cs.CV cs.LG
keywords fakemodelsimagesgenerativeclassificationdetectimageupon
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original abstract

With generative models proliferating at a rapid rate, there is a growing need for general purpose fake image detectors. In this work, we first show that the existing paradigm, which consists of training a deep network for real-vs-fake classification, fails to detect fake images from newer breeds of generative models when trained to detect GAN fake images. Upon analysis, we find that the resulting classifier is asymmetrically tuned to detect patterns that make an image fake. The real class becomes a sink class holding anything that is not fake, including generated images from models not accessible during training. Building upon this discovery, we propose to perform real-vs-fake classification without learning; i.e., using a feature space not explicitly trained to distinguish real from fake images. We use nearest neighbor and linear probing as instantiations of this idea. When given access to the feature space of a large pretrained vision-language model, the very simple baseline of nearest neighbor classification has surprisingly good generalization ability in detecting fake images from a wide variety of generative models; e.g., it improves upon the SoTA by +15.07 mAP and +25.90% acc when tested on unseen diffusion and autoregressive models.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 11 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. TextRich: A Multi-Domain Benchmark for Detecting AI-Generated Text-Rich Images from GPT-Image-2

    cs.CV 2026-06 unverdicted novelty 7.0

    Introduces a multi-domain benchmark for detecting AI-generated text-rich images from GPT-Image-2 and evaluates five detectors showing domain-dependent performance and JPEG sensitivity.

  2. Forged Calamity: Benchmark for Cross-Domain Synthetic Disaster Detection in the Age of Diffusion

    cs.CV 2026-06 unverdicted novelty 7.0

    Introduces Forged Calamity benchmark and shows that fine-tuned and zero-shot synthetic image detectors lose substantial accuracy on unseen generators and disaster types.

  3. GenSyn10: A Multi-Generative AI Dataset For Benchmarking Image Classification

    cs.CV 2026-07 conditional novelty 6.0

    GenSyn10 provides 60k CIFAR-10-aligned images from FLUX.2, HunyuanImage-3.0, and Qwen-Image-2512, showing detectors lose 4–18 points of accuracy on an unseen generator.

  4. TextRich: A Multi-Domain Benchmark for Detecting AI-Generated Text-Rich Images from GPT-Image-2

    cs.CV 2026-06 conditional novelty 6.0

    A new six-domain benchmark shows existing AI-image detectors are highly inconsistent on text-rich images and fail badly under JPEG compression, while a vision-language model is stronger but still weak on tables.

  5. When Eyes Betray AI: Social Gaze Consistency as a Semantic Cue for AI-Generated Image Detection

    cs.CV 2026-05 unverdicted novelty 6.0

    Social gaze consistency between interacting people is proposed as a new semantic cue orthogonal to low-level artifacts for detecting AI-generated images, with reported accuracy gains on vision and vision-language models.

  6. Deepfake Detection Generalization with Diffusion Noise

    cs.CV 2026-04 unverdicted novelty 6.0

    ANL uses diffusion noise prediction and attention to regularize deepfake detectors for better generalization to unseen synthesis methods without added inference cost.

  7. Navigating the Challenges of AI-Generated Image Detection in the Wild: What Truly Matters?

    cs.CV 2025-07 conditional novelty 6.0

    The ITW-SM dataset and targeted optimization of detector design choices yield a 26.87% average AUC improvement for state-of-the-art AI-generated image detectors under real-world social media conditions.

  8. Generative AI Literacy Training Improves Intelligence Analysts' Discrimination of Real and AI-Generated Images

    cs.HC 2026-06 conditional novelty 4.0

    A 30-minute training intervention increased US intelligence analysts' accuracy at distinguishing real from AI-generated images by 9 percentage points from a 72% baseline, mainly by improving identification of real images.

  9. Findings of the Counter Turing Test: AI-Generated Image Detection

    cs.CV 2026-05 unverdicted novelty 4.0

    A competition using a new 50k-image dataset found high accuracy in binary real-vs-AI detection but only modest success in identifying the exact generative model.

  10. Findings of the Counter Turing Test: AI-Generated Image Detection

    cs.CV 2026-05 unverdicted novelty 4.0

    The Counter Turing Test competition finds F1-scores above 0.83 for binary real-vs-AI classification but only 0.4986 at best for identifying the specific generative model.

  11. Findings of the Counter Turing Test: AI-Generated Image Detection

    cs.CV 2026-05 unverdicted novelty 3.0

    Binary AI vs. real image classification reaches F1 > 0.83 while identifying the exact generative model achieves a highest F1 of 0.4986 on the MS COCOAI dataset.