DailyBench: A Unified Benchmark for AI-Generated and Manipulated Images from Modern Generative Models

Researchers released DailyBench, a unified benchmark for AI image detection that covers both modern synthetic images and object-level edits on real photos.

It combines FakeBench for images from recent generative models and ManipulationBench for subtle, object-level manipulations.

Tests found that detectors that scored well on older benchmarks lose substantial accuracy on these more realistic and difficult cases.

The results show current evaluation sets are outdated and that more robust, manipulation-aware detection methods are needed.