AI-Safeguard/Ivy-Fake

VISIONPricing:Input $0.32 / Cached $0.016 / Output $1.6Concurrent Unit Cost:1Model Size:3BQuant:BF16Context Size:32kPublished:Jun 14, 2025Architecture:Transformer0.0K Featherless Exclusive Cold

AI-Safeguard/Ivy-Fake is a 3 billion parameter multimodal model developed by AI-Safeguard, specifically designed for detecting AI-generated images and videos. This model, based on the Qwen2.5-VL architecture, offers a unified and explainable framework for identifying synthetic media. Its primary use case is robust detection of deepfakes and other AI-generated content, providing reasoning for its classifications.

Loading preview...

Ivy-Fake: Unified Explainable AIGC Detection

AI-Safeguard/Ivy-Fake is a 3 billion parameter multimodal model, part of a unified framework for detecting AI-generated images and videos. Developed by AI-Safeguard, this model is based on the Qwen2.5-VL architecture and was accepted at ICMR 2026. It provides an explainable approach to identifying synthetic media, offering reasoning alongside its classifications.

Key Capabilities

  • Multimodal AIGC Detection: Specialized in identifying both AI-generated images and videos.
  • Explainable Framework: Provides reasoning for its detection classifications, enhancing transparency and trustworthiness.
  • Unified Approach: Offers a single model for diverse AI-generated content detection tasks.
  • Dedicated Dataset: Supported by the Ivy-Fake Dataset, released in December 2025, for robust training and evaluation.

Good For

  • Deepfake Detection: Identifying manipulated or synthetically generated visual media.
  • Content Authenticity Verification: Assessing whether images or videos are real or AI-generated.
  • Research in AIGC Forensics: Providing a benchmark and framework for studying AI-generated content detection.
  • Applications Requiring Explainable AI: Systems where understanding the 'why' behind a detection is crucial.

Finetunes
1 models