Original title: Multi-Tool Image Editing Attribution in Facial Forgery
Sheng Liu, Qiang Sheng, Danding Wang, Yu Li +2 · September 2026
Generative AI makes editing face images easy. However, finding which tools modified an image is hard when someone uses several tools in a row. Different tools leave overlapping traces that confuse older detection systems, which assume an editor used only one tool. To solve this, researchers studied Multi-Tool Image Editing Attribution, which aims to identify every tool used on a face image. They created MultiEdit, a dataset of 500k+ edited facial images featuring six types of tools for face swapping and facial enhancements. Then, they built an attribution method called DPEC. It works by capturing local tool traces from both normal image space and frequency patterns, guided by an error-based learning strategy. In tests, DPEC outperformed nine other methods on facial images edited in at most five steps.
Imagine you taste a soup and need to list every ingredient used, but the flavors have merged into one. Normally, you might identify just the dominant spice. Here, the system is like a chef who can pick out each individual ingredient by examining the soup's texture, aroma, and taste separately, even after they've blended.
When several AI tools are used to alter a face, it's tough to trace each change. This matters for verifying image authenticity, which helps prevent misinformation and ensures trust in digital media.
The abstract doesn't specify how the method handles more than five editing steps or tools not included in the dataset, and it relies on simulated editing operations.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Multi-Tool Image Editing Attribution in Facial Forgery — Sheng Liu, Qiang Sheng, Danding Wang et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.