---
license: apache-2.0
tags:
- IFDL
- MLLM
library_name: transformers
pipeline_tag: image-text-to-text
---

# FakeShield Model Card

## Model details

**Model type:** FakeShield is a multi-modal framework for explainable image forgery detection and localization (IFDL). It integrates multi-modal large language models (MLLMs) to analyze manipulated images, generate tampered region masks, and provide explanations. The model includes weights for both DTE-FDM and MFLM components.

The model repository includes weights for both DTE-FDM and MFLM components.

**Model date:** fakeshield-v1-22b was trained in September 2024.

**Project page:** https://zhipeixu.github.io/projects/FakeShield/

**Paper:** [FakeShield: Explainable Image Forgery Detection and Localization via Multi-modal Large Language Models](https://huggingface.co/papers/2410.02761)

## License

FakeShield is licensed under the Apache 2.0 License.

**Where to send questions or comments about the model:**

[https://github.com/zhipeixu/FakeShield/issues](https://github.com/zhipeixu/FakeShield/issues)

## Intended use

**Primary intended uses:** The primary use of FakeShield is used for research on image forgery detection and localization, enhancing transparency in media forensics.

**Primary intended users:** The primary intended users of the model are researchers, forensic analysts, journalists, and AI practitioners in media forensics and misinformation detection.