Marca de Agua IA Invisible y Escáner Forense de Frecuencia
Inspecciona anillos Fourier, mallas de escalado, planos de ruido LSB, discontinuidades ELA y marcas de agua ultrasónicas de 16–24 kHz en tu navegador.
Arrastra una imagen o pista de audio
Compatible con JPEG, PNG, WebP, AVIF, TIFF, BMP, MP3, WAV, M4A, FLAC, OGG. Cuadrícula máx 1024×1024 (~8MB RAM).
Garantía Sin Servidor: Todos los cálculos se realizan localmente en tu navegador.
Controles de Transformada Rápida de Fourier 2D
Forensic Signal Metrics
Executive Forensic Triage: How to Detect AI Images & Invisible Watermarks
Instant answers synthesized for researchers, forensic analysts, and search engines verifying synthetic media integrity.
Is this image AI generated?
To verify if an image is AI generated, apply a Two-Layer Forensic Audit: First, inspect Layer 1 (Container Provenance) for C2PA JUMBF manifests, IPTC DigitalSourceType tags, and camera EXIF headers. Second, inspect Layer 2 (Signal Mathematics) using a 2D Fast Fourier Transform (2D FFT) to identify periodic lattice crosshairs from neural upscalers, concentric Fourier carrier rings from latent watermarks (SynthID), and flattened Shannon bit-entropy (< 0.95) in the Least Significant Bit (LSB) plane.
What is an invisible AI frequency watermark (SynthID)?
An invisible AI frequency watermark (e.g., Google DeepMind SynthID, Meta Stable Signature) is an imperceptible signal embedded directly into the mathematical coefficients of an image (Discrete Wavelet Transform sub-bands or Fourier coefficients) or audio waveform during synthesis. Unlike EXIF or C2PA metadata stored in file headers, frequency watermarks alter spatial and spectral relationships, surviving screenshots, cropping, lossy JPEG recompression, and color grading. In 2D FFT spectral analysis, they emerge as circular rings or harmonic halos around the DC center.
How can you detect AI images when metadata is stripped?
When metadata is erased by social networks or screenshotting, detect AI images through frequency-domain and steganographic inspection: (1) 2D FFT exposes high-frequency grid artifacts and starburst crosshairs caused by transposed convolutions and pixel-shuffle decoders; (2) LSB Bit-Plane Slicing calculates Shannon entropy on Bit 0—natural camera sensors exhibit chaotic Poisson noise (~0.98–1.00), while AI generators leave flat zero-entropy dead zones; (3) Error Level Analysis (ELA) exposes localized recompression error disparities caused by AI inpainting or Generative Fill.
Container Provenance (Layer 1) vs Signal Forensics (Layer 2)
Why inspecting metadata alone is no longer enough to detect modern generative AI images and synthetic audio.
| Inspection Dimension | FreeImGen AI Signal Scanner | Layer 1 Metadata (C2PA/EXIF) | Traditional ELA Tools |
|---|---|---|---|
| Google SynthID Detection | 2D FFT Frequency Rings | Invisible to headers | Cannot detect |
| Survives Screenshot / Re-save | Embedded in pixel math | 100% stripped | ~ Degraded by re-save |
| Diffusion Upscaler Artifacts | Crosshairs & harmonic grids | No signal | ~ Inconclusive |
| Suno / Udio Ultrasonic Watermarks | 16–24 kHz STFT Waterfall | Missing in MP3s | Images only |
| Zero-Upload Privacy | Pure Client-Side TS | Local parser | Most upload to cloud |
AI Provenance & Watermark Ecosystem Tools
Seamlessly transition between metadata container inspection, C2PA manifest removal, and signal-level frequency forensics.
AI Metadata Checker
Inspect C2PA JUMBF containers, IPTC DigitalSourceType, and camera EXIF headers before running frequency analysis.
Remove Content Credentials
Sanitize C2PA cryptographic manifests and Adobe Content Credentials without re-encoding or degrading pixel quality.
Instagram AI Info Label Remover
Strip false-flag metadata tags triggering Meta automated "AI Info" badges on genuine photos edited in Lightroom.
Why Instagram Says AI Info on Real Photos
Read the comprehensive investigation explaining how phone camera AI and subtle retouching trigger automated platform warnings.
Profundización técnica: Esteganografía en el dominio de la frecuencia
Algoritmo 2D FFT Cooley-Tukey
Descomposición matemática de píxeles en frecuencias armónicas espaciales con memoria acotada a ~8MB.
Preguntas Frecuentes sobre Marcas de Agua IA Invisibles
¿Qué diferencia hay entre C2PA y SynthID?
C2PA es un metadato en la cabecera que se pierde con una captura de pantalla. SynthID está incrustado directamente en las frecuencias matemáticas de los píxeles o del sonido.
¿Se suben mis archivos a internet?
No. El análisis se ejecuta 100% en local en tu navegador mediante Web Worker y TypeScript.