ai-generated-image-detection

A curated collection of the latest research and resources on AI-Generated Image and Video Detection.

[CVPR 2026] UniGenDet: A Unified Generative-Discriminative Framework for Co-Evolutionary Image Generation and Generated Image Detection

Code for the LOTA: Bit-Planes Guided AI-Generated Image Detection

「ICML 2026」Official PyTorch Code for "OmniAID: Decoupling Semantics and Artifacts for Universal AI-Generated Image Detection in the Wild".

[CVPR 2026] Towards Generalizable AI-Generated Image Detection via Image-Adaptive Prompt Learning

Zero-shot AI-generated image detection using forensic self-descriptions. Trained only on real images, detects deepfakes from any generator. Official implementation. CVPR 2025.

SpecXNet is a robust AI-generated image and deepfake detection model that fuses spatial and frequency-domain features using Fourier-based attention for improved cross-generator generalization.

[ICCV 2025] D^3QE: Learning Discrete Distribution Discrepancy-aware Quantization Error for Autoregressive-Generated Image Detection

Official repository of the paper: Can ChatGPT Detect DeepFakes? A Study of Using Multimodal Large Language Models for Media Forensics

[arXiv 2026] "Micro-Defects expose Macro-Fakes: Detecting AI-Generated Images via Local Distributional Shifts"

Detects AI-generated faces (GAN, diffusion) with a MobileNetV2 transfer-learning classifier in PyTorch, about 95% accuracy on the 140k real-vs-fake faces dataset. Includes a FastAPI REST API, drag-and-drop web UI and a Colab/Kaggle training script; the model is published on Kaggle.

[CVPR Findings 2026] PRADA: Probability-Ratio-Based Attribution and Detection of Autoregressive-Generated Images

Research reference implementation of patch-wise supervision for AI-generated image detection.

Dual-stream spectral-spatial neural network for detecting AI-generated images (CIFAKE). 95.71% accuracy, 13.2x smaller than ResNet-18, with a full two-run replication that refutes the project's own central hypothesis.

SynthFlag — TikTok TechJam 2026 project for robust AI-generated image detection with a four-expert CLIP/SigLIP ensemble and evidence-aware reporting.

Digital media fingerprinting and source identification: detecting and attributing AI-generated images (B.Tech minor project, MNNIT Allahabad)

AI-powered food image verification system that detects food, classifies food categories, and identifies potentially AI-generated images using computer vision.

Multi-agent AI image forensics pipeline that detects AI-generated and manipulated images using parallel vision agents (Gemini + Groq) with a context-aware pre-processing layer and Cerebras-based arbitration to reduce false positives.

Nothing hides from Viy's gaze. Datasets, robust augmentations and benchmarks for AI-generated image detection.

DeForge-AI: A robust framework for AI-generated image detection, combining a DINOv3 Vision Transformer with high-frequency forensic noise analysis.