Research map / Image generation / Autoregressive and latent generation
Generalizable fake image detection
Detect AI-generated images with large-scale datasets, multi-generator training, and semantic-pixel features.
119 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2019–2026) · an approach in Autoregressive and latent generation, keyword cluster “ai · watermarking · generated”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
28 papers in 2023–24 and 80 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 5.9% to 8.9%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2019 · Unsupervised Fake News Detection on Social Media: A Generative Approach (AAAI 2019 · 310 citations)
- 2020 · CNN-Generated Images Are Surprisingly Easy to Spot... for Now (CVPR 2020 · 1,121 citations)
- 2021 · A Closer Look at Fourier Spectrum Discrepancies for CNN-Generated Images Detection (CVPR 2021 · 81 citations)
- 2022 · Proactive Image Manipulation Detection (CVPR 2022 · 44 citations)
- 2023 · Towards Universal Fake Image Detectors That Generalize Across Generative Models (CVPR 2023 · 381 citations)
- 2024 · Forgery-aware Adaptive Transformer for Generalizable Synthetic Image Detection (CVPR 2024 · 112 citations)
- 2025 · SIDA: Social Media Image Deepfake Detection, Localization and Explanation with Large Multimodal Model (CVPR 2025 · 53 citations)
- 2026 · Zooming In on Fakes: A Novel Dataset for Localized AI-Generated Image Detection with Forgery Amplification Approach (AAAI 2026 · 3 citations)
Earliest papers here
- Unsupervised Fake News Detection on Social Media: A Generative Approach (AAAI 2019 · 310 citations)
- AIRD: Adversarial Learning Framework for Image Repurposing Detection (CVPR 2019 · 29 citations)
- CNN-Generated Images Are Surprisingly Easy to Spot... for Now (CVPR 2020 · 1,121 citations)
What has been tried
67 papers here report 692 measured changes (a component added, removed or swapped, compared with a baseline): helped 59%, no clear effect 7%, hurt 28%, mixed 5%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add frequency-domain features to the forgery-aware adapter | GANs dataset, 4-class ProGAN; mean · ACCM | 95.4 -> 98.4; over −3.0% ACC gaps | Helped | Forgery-aware Adaptive Transformer for Generalizable Synthetic Image Detection (CVPR 2024 · 112 citations) |
| add an attention module between detection and segmentation features | SID-Set tampered localization · AUC | 84.1 -> 87.3 | Helped | SIDA: Social Media Image Deepfake Detection, Localization and Explanation with Large Multimodal Model (CVPR 2025 · 53 citations) |
| Train the diffusion model with timestep T=1 instead of the selected T=4 | tumor synthesis · segmentation performance | Hurt | Towards Generalizable Tumor Synthesis (CVPR 2024 · 36 citations) | |
| Remove text input from LGD | Cornell (H) · success rate | 0.86 -> 0.85 | Hurt | Language-driven Grasp Detection (CVPR 2024 · 34 citations) |
| use self-conditioned image generation rather than text-driven pairing for training data | Synthbuster and WildRF (AVG) · balanced accuracy | 61.9 -> 81.4 | Helped | A Bias-Free Training Paradigm for More General AI-generated Image Detection (CVPR 2025 · 24 citations) |
| replace MFM ViT-B/16 with CLIP ViT-B/16 as the backbone | 5 real-image sources, 13 generative models · AUC | 91.0 -> 87.6 | Hurt | Any-Resolution AI-Generated Image Detection by Spectral Learning (CVPR 2025 · 23 citations) |
| Increase the training set size | AI-Face test set · AUC | Helped | AI-Face: A Million-Scale Demographically Annotated AI-Generated Face Dataset and Fairness Benchmark (CVPR 2025 · 15 citations) | |
| Use CLIP ViT-L/14 rather than the other tested backbones | Stable Diffusion v1.4 training and generative-model tests · AP | Helped | Towards Universal AI-Generated Image Detection by Variational Information Bottleneck Network (CVPR 2025 · 14 citations) | |
| Train on more latent diffusion models while fixing the total number of training images | Extended evaluation set across generator types · mAP | Helped | Community Forensics: Using Thousands of Generators to Train Fake Image Detectors (CVPR 2025 · 10 citations) | |
| Use the patch-based classifier with zero padding instead of resizing | Open-world · Avg. Acc. | 82.12 -> 85.97 | Helped | Breaking Semantic Artifacts for Generalized AI-generated Image Detection (NeurIPS 2024 · 10 citations) |
| Add ODA | 17 GAN datasets, quality-aware, 2-class training · mean accuracy | 69.4 -> 71.0 | Helped | ODDN: Addressing Unpaired Data Challenges in Open-World Deepfake Detection on Online Social Networks (AAAI 2025 · 7 citations) |
| Use adaptive fusion instead of straightforward feature-integration methods | Synthetic image detection · detection performance | Helped | CO-SPY: Combining Semantic and Pixel Features to Detect Synthetic Images by AI (CVPR 2025 · 7 citations) |
What reviewers will expect
Datasets and baselines named in the 39 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- GenImage · 12 of 39 papers
- ImageNet · 9 of 39 papers
- COCO · 8 of 39 papers
- ForenSynths · 5 of 39 papers
- DRCT-2M · 5 of 39 papers
- LSUN · 5 of 39 papers
- LAION · 4 of 39 papers
- FFHQ · 4 of 39 papers
Baselines
- NPR · 16 of 39 papers
- UnivFD · 13 of 39 papers
- DIRE · 12 of 39 papers
- LGrad · 12 of 39 papers
- FatFormer · 9 of 39 papers
- CNNSpot · 9 of 39 papers
- DRCT · 9 of 39 papers
- F3Net · 7 of 39 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 108 papers; from OpenAlex and the papers' affiliations).
- University of Science and Technology of China · 5 papers
- Peking University · 5 papers
- Tencent · 4 papers
- Beijing Jiaotong University · 3 papers
- Shanghai Jiao Tong University · 3 papers
- Baidu · 3 papers
- Rutgers University · 3 papers
- Sony · 3 papers
- Stanford University · 3 papers
- Google · 3 papers
Use it
Download BibTeX of 33 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Robust diffusion watermarking (79 papers)
- Reconstruction-based diffusion image detection (45 papers)
Elsewhere in the topic
- Classical generative image synthesis (147 papers)
- High-resolution diffusion synthesis (140 papers)
- Retrieval and matching with generative features (140 papers)
- Generative latent space tooling (129 papers)
- Latent diffusion with learned tokenizers (114 papers)
- Synthetic training data from generators (111 papers)
- Compositional scene generation (105 papers)
- Next-scale and token autoregressive images (90 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).