With the increasing popularity of AI-generated content, how to effectively identify its authenticity has become an important issue. In order to solve this problem, Tencent launched an AI-generated image identification tool on January 17, 2025. This tool was developed by the Zhuque Lab of Tencent's Hunyuan security team. It is designed to help users quickly and accurately identify whether images are generated by AI, improve content identification efficiency, and maintain the authenticity and security of the network environment. This tool uses an advanced AI model to make judgments by analyzing multiple dimensions such as texture, semantics and invisible features of the image, effectively improving the detection accuracy.
On January 17, 2025, Tencent launched an AI-generated image identification tool designed to help people identify whether images were generated by AI.
The AI-generated image detection system developed by the Zhuque Laboratory of Tencent's Hunyuan Security Team can determine whether the image was generated by AI in a few seconds by uploading the image and waiting for verification. The system mainly distinguishes by capturing the difference between real pictures and AI-generated pictures. For example, AI-generated pictures sometimes do not conform to common sense logic, and there will be content that does not conform to reality, such as puppies flying with wings and cats holding cigars. AI-generated pictures need to be "watermarked", that is, add explicit Or implicit signs. These signs may be visible to the naked eye or may not be obvious and need to be read with detection tools; AI-generated images also contain some hidden features that are invisible to the naked eye. For example, after adjusting the HSV color space, the AI-generated images will appear. Highlights with densely distributed local textures, etc.
Identifying images generated by AI cannot rely solely on a single basis. The detection system needs to use AI models to capture differences in various features, including the texture, semantics and invisible features of the image. Suzaku Lab used 1.4 million positive and negative samples for model training, covering a variety of content generation scenarios. The final test detection rate reached more than 95%, and it is still being optimized and improved. Suzaku Lab has also developed a text detection system, which implements text detection by learning from massive data. It also collects a large number of positive and negative samples for training, and uses comparative methods to infer the AI generation probability of articles to enhance the detection of unseen data. ability. At present, the system covers diverse literary genres such as newsletters, official documents, novels, and essays, and will be enhanced for poetry and other genres in the future.
The original intention of the Suzaku Laboratory in developing AI detection tools was to optimize the training corpus of Hunyuan large models. However, with the explosive growth of AI, the demand for "using AI to detect AI generation" increased in academia and industry, so the laboratory developed AI generation There are two tools for image detection and AI-generated text detection, and experience services are provided.
In the future, detection tools for AI-generated videos will also be launched to further improve detection capabilities.
Experience address:
https://matrix.tencent.com/ai-detect/ai_gen
https://matrix.tencent.com/ai-detect
The launch of Tencent's AI-generated image identification tool marks important progress in combating false information in AI-generated content. In the future, as technology continues to improve, I believe that such tools will better serve society and maintain the healthy development of the network ecosystem.