Synthetic Image Detection with CLIP: Understanding and Assessing Predictive Cues

• URL: http://arxiv.org/abs/2602.12381v1
• Date: Thu, 12 Feb 2026 20:21:32 GMT
• Title: Synthetic Image Detection with CLIP: Understanding and Assessing Predictive Cues
• Abstract summary: It is unclear whether CLIP-based detectors simply detect strong visual artifacts or exploit subtle semantic biases.<n>We analyze what CLIP-based detectors learn using an interpretable linear head with de-correlated activations and a text-grounded concept-model.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Recent generative models produce near-photorealistic images, challenging the trustworthiness of photographs. Synthetic image detection (SID) has thus become an important area of research. Prior work has highlighted how synthetic images differ from real photographs--unfortunately, SID methods often struggle to generalize to novel generative models and often perform poorly in practical settings. CLIP, a foundational vision-language model which yields semantically rich image-text embeddings, shows strong accuracy and generalization for SID. Yet, the underlying relevant cues embedded in CLIP-features remain unknown. It is unclear, whether CLIP-based detectors simply detect strong visual artifacts or exploit subtle semantic biases, both of which would render them useless in practical settings or on generative models of high quality. We introduce SynthCLIC, a paired dataset of real photographs and high-quality synthetic counterparts from recent diffusion models, designed to reduce semantic bias in SID. Using an interpretable linear head with de-correlated activations and a text-grounded concept-model, we analyze what CLIP-based detectors learn. CLIP-based linear detectors reach 0.96 mAP on a GAN-based benchmark but only 0.92 on our high-quality diffusion dataset SynthCLIC, and generalization across generator families drops to as low as 0.37 mAP. We find that the detectors primarily rely on high-level photographic attributes (e.g., minimalist style, lens flare, or depth layering), rather than overt generator-specific artifacts. CLIP-based detectors perform well overall but generalize unevenly across diverse generative architectures. This highlights the need for continual model updates and broader training exposure, while reinforcing CLIP-based approaches as a strong foundation for more universal, robust SID.

Related papers

• The Alpha Blending Hypothesis: Compositing Shortcut in Deepfake Detection [48.85953407706351]
  We introduce the Alpha Blending Hypothesis, positing that state-of-the-art frame-based detectors primarily function as alpha blending searchers.<n>We experimentally validate the hypothesis, demonstrating that deepfake detectors exhibit high sensitivity to the so-called self-blended images.<n>We show that predictions from explicit blending searchers and models resilient to blending shortcuts are highly complementary, yielding a state-of-the-art AUROC of 94.0% in an ensemble configuration.
  arXiv  Detail & Related papers  (2026-05-11T10:35:39Z)
• Beyond Semantics: Uncovering the Physics of Fakes via Universal Physical Descriptors for Cross-Modal Synthetic Detection [34.95491424781195]
  We propose a novel feature selection algorithm that identifies five core physical features that exhibit consistent discnative power across all tested datasets.<n>Our method achieves state-of-the-art performance on multiple Genimage benchmarks, with near-perfect accuracy (99.8%) on datasets such as Wukong and SDv1.4.
  arXiv  Detail & Related papers  (2026-04-06T11:50:29Z)
• Layer Consistency Matters: Elegant Latent Transition Discrepancy for Generalizable Synthetic Image Detection [15.935593506859043]
  Real images maintain consistent semantic attention and structural coherence in their latent representations, whereas synthetic ones present discernible patterns.<n>We propose a novel approach termed latent transition discrepancy (LTD), which captures the inter-layer consistency differences of real and synthetic images.<n>Our approach exceeds the base model by 14.35% in mean Acc across three datasets containing diverse GANs and DMs.
  arXiv  Detail & Related papers  (2026-03-11T09:53:06Z)
• Detecting AI-Generated Images via Distributional Deviations from Real Images [6.615773227400183]
  We propose a Masking-based Pre-trained model Fine-Tuning (MPFT) strategy, which introduces a Texture-Aware Masking (TAM) mechanism to mask textured areas containing generative model-specific patterns during fine-tuning.<n>Our method, fine-tuned with only a minimal number of images, significantly outperforms existing approaches, achieving up to 98.2% and 94.6% average accuracy on the two datasets, respectively.
  arXiv  Detail & Related papers  (2026-01-07T05:00:13Z)
• AdaptPrompt: Parameter-Efficient Adaptation of VLMs for Generalizable Deepfake Detection [7.76090543025328]
  Recent advances in image generation have led to the widespread availability of highly realistic synthetic media, increasing the difficulty of reliable deepfake detection.<n>A key challenge is generalization, as detectors trained on a narrow class of generators often fail when confronted with unseen models.<n>We address the pressing need for generalizable detection by leveraging large vision-language models, specifically CLIP, to identify synthetic content across diverse generative techniques.
  arXiv  Detail & Related papers  (2025-12-19T16:06:03Z)
• Self-Supervised AI-Generated Image Detection: A Camera Metadata Perspective [80.10217707456046]
  We introduce a self-supervised approach for detecting AI-generated images that leverages camera metadata.<n>We train a feature extractor solely on camera-captured photographs by classifying categorical EXIF tags.<n>Our detectors deliver strong generalization to in-the-wild samples and robustness to common benign image perturbations.
  arXiv  Detail & Related papers  (2025-12-05T11:53:18Z)
• NS-Net: Decoupling CLIP Semantic Information through NULL-Space for Generalizable AI-Generated Image Detection [14.7077339945096]
  NS-Net is a novel framework that decouples semantic information from CLIP's visual features, followed by contrastive learning to capture intrinsic distributional differences between real and generated images.<n>Experiments show that NS-Net outperforms existing state-of-the-art methods, achieving a 7.4% improvement in detection accuracy.
  arXiv  Detail & Related papers  (2025-08-02T07:58:15Z)
• DeeCLIP: A Robust and Generalizable Transformer-Based Framework for Detecting AI-Generated Images [14.448350657613368]
  DeeCLIP is a novel framework for detecting AI-generated images.<n>It incorporates DeeFuser, a fusion module that combines high-level and low-level features.<n>We trained exclusively on 4-class ProGAN data, DeeCLIP achieves an average accuracy of 89.90%.
  arXiv  Detail & Related papers  (2025-04-28T15:06:28Z)
• FakeScope: Large Multimodal Expert Model for Transparent AI-Generated Image Forensics [66.14786900470158]
  We propose FakeScope, an expert multimodal model (LMM) tailored for AI-generated image forensics.<n>FakeScope identifies AI-synthetic images with high accuracy and provides rich, interpretable, and query-driven forensic insights.<n>FakeScope achieves state-of-the-art performance in both closed-ended and open-ended forensic scenarios.
  arXiv  Detail & Related papers  (2025-03-31T16:12:48Z)
• CO-SPY: Combining Semantic and Pixel Features to Detect Synthetic Images by AI [58.35348718345307]
  Current efforts to distinguish between real and AI-generated images may lack generalization.<n>We propose a novel framework, Co-Spy, that first enhances existing semantic features.<n>We also create Co-Spy-Bench, a comprehensive dataset comprising 5 real image datasets and 22 state-of-the-art generative models.
  arXiv  Detail & Related papers  (2025-03-24T01:59:29Z)
• Understanding and Improving Training-Free AI-Generated Image Detections with Vision Foundation Models [68.90917438865078]
  Deepfake techniques for facial synthesis and editing pose serious risks for generative models.<n>In this paper, we investigate how detection performance varies across model backbones, types, and datasets.<n>We introduce Contrastive Blur, which enhances performance on facial images, and MINDER, which addresses noise type bias, balancing performance across domains.
  arXiv  Detail & Related papers  (2024-11-28T13:04:45Z)
• Contrasting Deepfakes Diffusion via Contrastive Learning and Global-Local Similarities [88.398085358514]
  Contrastive Deepfake Embeddings (CoDE) is a novel embedding space specifically designed for deepfake detection. CoDE is trained via contrastive learning by additionally enforcing global-local similarities.
  arXiv  Detail & Related papers  (2024-07-29T18:00:10Z)
• Parents and Children: Distinguishing Multimodal DeepFakes from Natural Images [60.34381768479834]
  Recent advancements in diffusion models have enabled the generation of realistic deepfakes from textual prompts in natural language. We pioneer a systematic study on deepfake detection generated by state-of-the-art diffusion models.
  arXiv  Detail & Related papers  (2023-04-02T10:25:09Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.