Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Unknown Aware AI-Generated Content Attribution

Ellie Thieu, Jifan Zhang, Haoyue Bai

TL;DR

This work targets target generator attribution in open-world settings, addressing the challenge of unseen and rapidly evolving generative models. It introduces an unknown-aware attribution framework that first trains a CLIP-based, linear classifier on labeled in-distribution data and then fine-tunes with unlabeled wild data under a constraint that preserves in-distribution performance. A Lagrangian objective balances exposure to diverse wild samples with safeguarding ID accuracy, yielding improved attribution for challenging, unseen generators such as Midjourney, Firefly, and Stable Diffusion XL while maintaining performance on known sources. Experiments demonstrate stability across wild-data sizes and robustness to label noise, suggesting practical deployment benefits where labeled data are scarce and model landscapes continually evolve.

Abstract

The rapid advancement of photorealistic generative models has made it increasingly important to attribute the origin of synthetic content, moving beyond binary real or fake detection toward identifying the specific model that produced a given image. We study the problem of distinguishing outputs from a target generative model (e.g., OpenAI Dalle 3) from other sources, including real images and images generated by a wide range of alternative models. Using CLIP features and a simple linear classifier, shown to be effective in prior work, we establish a strong baseline for target generator attribution using only limited labeled data from the target model and a small number of known generators. However, this baseline struggles to generalize to harder, unseen, and newly released generators. To address this limitation, we propose a constrained optimization approach that leverages unlabeled wild data, consisting of images collected from the Internet that may include real images, outputs from unknown generators, or even samples from the target model itself. The proposed method encourages wild samples to be classified as non target while explicitly constraining performance on labeled data to remain high. Experimental results show that incorporating wild data substantially improves attribution performance on challenging unseen generators, demonstrating that unlabeled data from the wild can be effectively exploited to enhance AI generated content attribution in open world settings.

Unknown Aware AI-Generated Content Attribution

TL;DR

This work targets target generator attribution in open-world settings, addressing the challenge of unseen and rapidly evolving generative models. It introduces an unknown-aware attribution framework that first trains a CLIP-based, linear classifier on labeled in-distribution data and then fine-tunes with unlabeled wild data under a constraint that preserves in-distribution performance. A Lagrangian objective balances exposure to diverse wild samples with safeguarding ID accuracy, yielding improved attribution for challenging, unseen generators such as Midjourney, Firefly, and Stable Diffusion XL while maintaining performance on known sources. Experiments demonstrate stability across wild-data sizes and robustness to label noise, suggesting practical deployment benefits where labeled data are scarce and model landscapes continually evolve.

Abstract

The rapid advancement of photorealistic generative models has made it increasingly important to attribute the origin of synthetic content, moving beyond binary real or fake detection toward identifying the specific model that produced a given image. We study the problem of distinguishing outputs from a target generative model (e.g., OpenAI Dalle 3) from other sources, including real images and images generated by a wide range of alternative models. Using CLIP features and a simple linear classifier, shown to be effective in prior work, we establish a strong baseline for target generator attribution using only limited labeled data from the target model and a small number of known generators. However, this baseline struggles to generalize to harder, unseen, and newly released generators. To address this limitation, we propose a constrained optimization approach that leverages unlabeled wild data, consisting of images collected from the Internet that may include real images, outputs from unknown generators, or even samples from the target model itself. The proposed method encourages wild samples to be classified as non target while explicitly constraining performance on labeled data to remain high. Experimental results show that incorporating wild data substantially improves attribution performance on challenging unseen generators, demonstrating that unlabeled data from the wild can be effectively exploited to enhance AI generated content attribution in open world settings.
Paper Structure (31 sections, 3 equations, 1 figure, 1 table)

This paper contains 31 sections, 3 equations, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Synthetic Image Generation
  4. AI-Generated Content Detection and Attribution
  5. Open-Set Attribution
  6. Semi-Supervised Learning
  7. Problem Setup
  8. Labeled in-distribution data.
  9. Unlabeled wild data.
  10. Learning framework.
  11. Unknown-Aware AI-Generated Content Attribution
  12. Training on In-Distribution Data
  13. Fine-Tuning with Wild Data via Constrained Optimization
  14. Learning Objective
  15. Experiments
  16. ...and 16 more sections

Figures (1)

  • Figure 1: Illustration of unknown-aware generator attribution. A classifier trained only on labeled sources learns a decision boundary that separates the target generator from known sources but may misclassify samples from unknown or unseen generators. By incorporating unlabeled wild data under a constrained optimization framework, the decision boundary is adjusted to better separate the target generator from diverse unknown sources, while preserving performance on labeled source data.