IRET is a deepfake and AI-generated media detection service built for the MENA region, with first-class support for Arabic and English content.
IRET analyzes images, videos, and audio files and determines whether they have been manipulated by AI or generated synthetically. It returns a clear verdict, a confidence score, and a human-readable explanation in both Arabic and English.
The system is designed to handle media as it actually circulates in the real world — including files shared through WhatsApp, Telegram, and other social platforms, where re-encoding and compression have stripped away most original metadata.
| Type | Formats |
|---|---|
| Image | JPG, PNG, BMP, WebP, TIFF |
| Video | MP4, AVI, MOV, MKV, WebM |
| Audio | MP3, WAV, FLAC, M4A, AAC, OGG |
- Image deepfake detection — face swaps, AI-generated portraits, GAN and diffusion model outputs
- Video deepfake detection — frame-level and temporal analysis, face manipulation, audio-visual consistency
- Audio manipulation detection — voice cloning, synthetic speech, audio splicing
- Social-media-aware analysis — automatically adjusts its detection strategy when files show signs of WhatsApp/Telegram re-encoding
- Bilingual output — all verdicts and explanations are delivered in both Arabic (MSA) and English
- Court-admissible PDF reports — structured forensic reports in Arabic suitable for legal proceedings
Every analysis returns one of the following verdicts:
| Verdict | Meaning |
|---|---|
REAL |
No evidence of AI generation or manipulation |
FAKE |
Strong evidence of AI manipulation or synthesis |
AI_GENERATED |
Content appears to be fully AI-generated (no original source) |
AUDIO_MANIPULATED |
Audio track shows signs of manipulation; visual may be authentic |
VIDEO_MANIPULATED |
Video track manipulated; audio may be authentic |
UNCERTAIN |
Insufficient signal — file may be too compressed, too short, or too low resolution for a reliable verdict |
The service exposes a REST API:
POST /detect Upload a file, receive verdict + explanation
POST /detect_frames Frame-by-frame heatmap analysis (image or video)
POST /api/report Generate an Arabic court-admissible PDF report
GET /health Service liveness check
GET /layers List active detection signals and their current weights
Authentication is required. Contact the IRET team for API access.
All user-facing output (verdicts, explanations, reports) is available in:
- Arabic — Modern Standard Arabic (MSA), suitable for formal and legal contexts
- English
- Legal and investigative teams in the MENA region requiring court-ready forensic documentation
- Media organizations verifying the authenticity of submitted content
- Journalists and fact-checkers working with Arabic-language media
- Developers and platforms needing an API for automated content moderation
IRET is developed by a team focused on digital trust and media integrity in the Arab world. It is the first deepfake detection system built specifically for the MENA region's media environment and legal requirements.
For inquiries, access requests, or partnerships, please reach out through the official IRET channels.