Multimodal AI models detect fake images by spotting physical flaws
Beyond Generation: Using Multimodal LLMs as Detectors of LLM-Generated Deepfake Facial Images
Computer Vision and Pattern RecognitionEmerging Technologies
Summary
Fake images are getting so realistic that usual ways to spot them aren’t working well. The authors tested several large multimodal AI models (MLLMs) to see if they can tell fake pictures from real ones by noticing physical mistakes instead of just hidden watermarks. They found some of these AI models, like GPT-5.2, can detect fakes with high accuracy by looking for those flaws. However, when images have invisible watermarks embedded, the detection gets a bit harder because the watermarking might hide some of the physical clues. This shows these AI models can help detect fake images but can be tricked by clever watermarking.
What this means in practice
- •For media verification teams: Use multimodal AI models to identify fake facial images by detecting structural flaws rather than relying on hidden watermarks.
- •For social media content moderators: Implement AI-driven tools to flag or filter synthetic images exploiting physical inconsistencies for better content trustworthiness.
Authors
Olivia Salamon, Virginia N. L. Franqueira, Shujun Li
Abstract
As image generation models become increasingly capable of producing highly realistic content, traditional forensic techniques are struggling to keep up. This paper investigates whether MLLMs can instead be used as a form of forensic analysis by identifying synthetic media. Specifically, the study aims to determine whether these models detect fake images by recognising physical inconsistencies, or whether they rely on digital artefacts such as embedded watermarks.Six MLLMs: GPT-4o, GPT-5, GPT-5.2, Gemini 2.5, Gemini 3, and Seed 1.6, are benchmarked using a dataset of 400 images. Fifty images are authentic photographs, while the remaining 350 are generated using seven leading image models: Gemini 3, Gemini 2.5, Flux.2 Pro, GPT-Image-1, GPT-Image-1.5, Seedream 4.0, and Seedream 4.5. The evaluation is structured in three stages: single-image classification, pairwise comparison, and a focused "self-detection" experiment on Gemini. Several MLLMs achieved a high detection accuracy, with GPT-5.2 reaching a peak of 97.9% in the pairwise experiment, against specific generators and an overall mean pairwise accuracy of 82.5% by identifying structural or physical inconsistencies within images. However, results from the self-detection experiment indicate that the presence of invisible SynthID watermarks leads to a 10% reduction in Gemini's detection accuracy compared to the clean control group. One possible explanation is that SynthID embedding reduces some of the visible artefacts the models rely on for detection.These results imply that MLLMs can detect structural artefacts without relying on watermarks, but performance varies substantially across model generations. The findings from comparison C5 also suggest that SynthID embedding may suppress some of the structural artefacts used during detection.