Prefill-Guided Thinking for zero-shot detection of AI-generated images
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915754111139840 |
|---|---|
| author | Kachwala, Zoher Singh, Danishjeet Yang, Danielle Menczer, Filippo |
| author_facet | Kachwala, Zoher Singh, Danishjeet Yang, Danielle Menczer, Filippo |
| contents | Traditional supervised methods for detecting AI-generated images depend on large, curated datasets for training and fail to generalize to novel, out-of-domain image generators. As an alternative, we explore pre-trained Vision-Language Models (VLMs) for zero-shot detection of AI-generated images. We evaluate VLM performance on three diverse benchmarks encompassing synthetic images of human faces, objects, and animals produced by 16 different state-of-the-art image generators. While off-the-shelf VLMs perform poorly on these datasets, we find that prefilling responses effectively guides their reasoning -- a method we call Prefill-Guided Thinking (PGT). In particular, prefilling a VLM response with the phrase "Let's examine the style and the synthesis artifacts" improves the Macro F1 scores of three widely used open-source VLMs by up to 24%. We analyze this improvement in detection by tracking answer confidence during response generation. For some models, prefills counteract early overconfidence -- akin to mitigating the Dunning-Kruger effect -- leading to better detection performance. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_11031 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Prefill-Guided Thinking for zero-shot detection of AI-generated images Kachwala, Zoher Singh, Danishjeet Yang, Danielle Menczer, Filippo Machine Learning Artificial Intelligence Computation and Language Traditional supervised methods for detecting AI-generated images depend on large, curated datasets for training and fail to generalize to novel, out-of-domain image generators. As an alternative, we explore pre-trained Vision-Language Models (VLMs) for zero-shot detection of AI-generated images. We evaluate VLM performance on three diverse benchmarks encompassing synthetic images of human faces, objects, and animals produced by 16 different state-of-the-art image generators. While off-the-shelf VLMs perform poorly on these datasets, we find that prefilling responses effectively guides their reasoning -- a method we call Prefill-Guided Thinking (PGT). In particular, prefilling a VLM response with the phrase "Let's examine the style and the synthesis artifacts" improves the Macro F1 scores of three widely used open-source VLMs by up to 24%. We analyze this improvement in detection by tracking answer confidence during response generation. For some models, prefills counteract early overconfidence -- akin to mitigating the Dunning-Kruger effect -- leading to better detection performance. |
| title | Prefill-Guided Thinking for zero-shot detection of AI-generated images |
| topic | Machine Learning Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2506.11031 |